RU2721178C1 - Intelligent automatic intruders detection system - Google Patents

Abstract

FIELD: security alarm.

SUBSTANCE: invention relates to security alarm and can be used in security systems for automatic detection of intruders who penetrated into protected territory and by their actions of triggering detection devices. Such result is achieved by intelligent automatic system for detecting intruders, comprising detection means in the form of security alarm sensors, connected to sensors logic processing unit and artificial neural network with training signals supplied to it, detection means in the form of imitators of intruders images, combined together with sensors of security alarm in N units of detection means, N units of devices for registration and primary processing of signals from detection means, N units of artificial neural networks performing recognition of signals from detection means, data flow control unit, unit of natural factors registration devices, a unit for means of recording parameters of anthropogenic factors, a unit for recording parameters of the external environment, a memory unit, a unit for final recognition and a decision unit.

EFFECT: technical result of the invention is to increase the probability of detecting unauthorized access to the protected territory and movement of violators of natural or artificial origin capable of endangering the secure facility.

13 cl, 2 dwg

Description

FIELD OF THE INVENTION

The invention relates to the field of security alarms and can be used in security systems for automatic (without human operator) detection of intruders (both single and group) who penetrated (trying to penetrate) into the protected area and by their actions trigger detection means. At the same time, a violator is not only a person who unauthorized (penetrates) into a protected area and creates a criminal threat to a protected object and (or) property, or a group of such people, but also any unauthorized persons (trying to enter) into a protected area that are capable of moving on the ground or at a height dangerous for the protected object over the protected territory and creating technical (artificial) means and complexes that create a criminal threat, for example, robotic (both autonomous, including with artificial intelligence, and controlled by the operator, including remotely), and also animals or birds that are capable (due to their preliminary purposeful training) of penetrating the protected object and creating the aforementioned threat. Those. a person creating a criminal threat and named in accordance with the current GOST (GOST R 52551-2016 “Security and safety systems. Terms and definitions”) as a “violator” (in legal terms) can not only penetrate the protected object (including using technical means or on technical means, for example, by car or gyroplane), but it can also send an autonomous (remotely controlled) technical tool (complex) or a trained animal (capable of moving and committing necessary illegal actions) to a guarded object to create a criminal threat bird). Trends in the development of robotics and artificial intelligence systems show that in the near future the bulk of penetration into protected areas with various (in particular, sabotage) targets will be carried out using autonomous and (or) remotely controlled (in particular, radio-controlled) technical means and complexes (for example, single copters (in particular, quadrocopters) or their groups). Such tools and complexes can commit unlawful actions under conditions in which, for various reasons, a person-violator cannot act (danger or incompatibility with a person’s life, mismatch of a person’s body size and physical capabilities required for penetration and commission of illegal actions, etc.). Therefore, it is advisable to refer to “violators” not only a person, but also all objects that have penetrated (trying to penetrate) into a protected area and (or) create a criminal threat (capable of creating a criminal threat) to a protected object, either on the instructions of a person or on their own solution (for technical objects with universal (general) artificial intelligence - a very real situation in the not-so-distant future). This is exactly the approach used in the invention.

BACKGROUND

The well-known "Intelligent alarm system with the ability to exchange information between detection tools" (patent RU 2594478, IPC G08B 13/00, priority from 06/16/2015). The system consists of a central guard post, a multitude of detection means (sensors) associated with the guard post and located on numbered sections of the protected area and operating on various physical principles (radio wave, radio beam, infrared, vibration (vibrometric), seismic, acoustic , magnetometric), and a digital weather station associated with a guard post and means of detection and allowing weather conditions to be taken into account, and between the central guard station and each of the detection means two-way exchange of information is carried out using the address principle.

Common with the claimed device signs: technical means of detection, operating on various physical principles and located on well-known areas of the protected area, as well as associated with the technical means of detection and between a central control point (central guard post) and a digital weather station.

The intellectuality of the proposed system, according to the authors, consists of the following possibilities: - taking into account weather conditions in order to adjust processing algorithms in detection tools (for example, during rain, hail, strong wind, fog, etc.); - interrogation of the state of neighboring detection means in order to make a final decision about the alarm situation at a certain section of the guard line (for example, during a thunderstorm or strong gusts of wind); - providing automatic or automated (without the participation of a central guard post) certain procedures for setting up detection tools (for example, adjusting the transmitting and receiving units in active on-off alarm means); - accumulation of data to take them into account when analyzing the state of detection tools when similar situations arise in the future; - analysis of the pre-alarming, alarming and post-alarming situation; - use of guard mode and sleep mode to save power by the system; - distribution and redistribution of computing resources between the central guard post and the first means of detection; - the use of neural network algorithms and fuzzy logic algorithms for intelligent signal processing.

However, the noted intelligence is not supported by the essential features in the claims. There are no examples of technical implementation of the above capabilities. Ensuring the achievement of the stated objective of the invention ("providing the ability to exchange information between the means of detection and use of a digital weather station to take into account weather conditions"), the structural diagram of the system and a description of its operation do not disclose how "exchange of information" and "accounting of weather conditions" are used to implement attributed by the authors to the manifestation of intelligence of the above system capabilities. In the description of the invention to the patent, unfortunately, there are neither schemes for implementing the above possibilities, nor links to sources of information with such schemes (both individually and (which is much more complicated) that do not interfere with each other). “Information exchange” is a too broad property that cannot be obviously correlated with the improvement of any consumer quality of the system considered by the authors (except for partial unloading of the central processor and communication channel). It should be noted that the technical implementation of most of the capabilities individually indicated by the authors that carry intellectuality individually (and even more so of their one or another combination) with the desired quality in itself is a complex and ambiguously solved scientific and technical problem that can make up the subject of individual inventions. Therefore, the authors' reference to the above-mentioned “intellectual” possibilities of the invention can be considered precisely as possibilities (as wishes not without foundation), but not as signs (properties) inherent in the invention. Moreover, the purpose of the invention ("providing the ability to exchange information between the means of detection and use of a digital weather station to take into account weather conditions") is clearly not related to the above "intelligent" capabilities.

The disadvantage of this system is the weak potential for the desirable (for making the right decision) classification of objects that trigger detection means (human, animal, bird, ground or flying technical device, ...), as well as the low ability to separate cases of detection means from these objects from cases of detection means triggering from interfering factors of natural (natural) and artificial (including targeted) origin. The reason is the insufficiency and uncertainty used for the detection and recognition of sensory information. The use of modern neural network algorithms and fuzzy logic algorithms in such conditions will not give the desired positive effect. The fact is that the use of modern artificial neural networks requires their lengthy preliminary training using large representative samples of the source data, as well as large time and computational costs, without any special guarantee of the effectiveness of further working use of neural networks in conditions of insufficiency and (or) noise ) sensory data. As you know, neural networks in the conditions of poor-quality training or lack of sensory information in the process of working, often when functioning in real conditions, show their "weaknesses" - they can be grossly mistaken, they are easy to fool, which is generally a consequence of the lack of ability of artificial neural networks to understand ”(everything that modern artificial neural networks do is just transform the input data vector into the output vector) as a result of their lack of one or another (depending on the area of application) spatio-temporal“ species ”and“ individual ”“ world models ” ”, Requiring an appropriate ontology, etc. In this case, this general shortcoming is exacerbated by the insufficiency and uncertainty used to detect and recognize sensory information. It should be noted that even the presence of a professional operator (natural intelligence) at the central guard post does not significantly improve the situation, because the information for making the right decision even by a professional person is clearly not enough.

The well-known "Intelligent Network System for Monitoring the Protected Territory" (patent RU 2629521, IPC G08B 13/00, priority dated June 17, 2017), claimed from the same as the above patent, is partially organized by the same team of authors. This system contains a central control center, detection equipment (alarm sensors), operating on different physical principles (seismic, radio wave, infrared, magnetometric, radio beam and breakaway), located on numbered sections of the protected area and connected with each other and with the central control center using the first radio channel with two-way radio communication, technical means of video surveillance (video surveillance system), interconnected with the central point volume control using a second radio channel with two-way radio. In this case, information transmission routes are automatically selected based on known methods of self-organization of data transmission networks according to the criterion of the best quality of radio communication using the relay capabilities of the technical means of detection and video surveillance.

The availability of technical means of video surveillance allows visual identification of a “person-intruder to distinguish him from animals, birds and other interference factors”.

Common with the claimed device are the signs: technical means of detection, operating on various physical principles and located on known areas of the protected territory, technical means of video surveillance, a central control center with a processor associated with the means of detection and video surveillance.

Information processing in the system, as stated in the description of the invention to the patent, is carried out with the possibility of using intelligent algorithms to achieve the maximum probability of detection and the minimum number of false alarms, taking into account the physical principle of operation of alarm sensors and the possibility of combining them, the location of the sensors on the ground, the current operating environment in the protected area, as well as the direction of movement of the detected object. Information processing algorithms are selected in accordance with the decisive rules “AND”, “OR”, “2 of 3” depending on tactical tasks. However, as in the previous patent, any technical implementation of the "intelligence" of the proposed system, demonstrating the ability to achieve the capabilities indicated in the patent:

- taking into account weather conditions in order to adjust processing algorithms (for example, during rain, hail, strong wind, fog, etc.);

- interrogation of the state of neighboring technical security equipment in order to make a final decision about the alarm situation at a certain section of the security line (for example, during a thunderstorm or strong gusts of wind);

- accumulation and storage of routing algorithms for use in possible options for the organization of radio networks in the future;

- accumulation of data to take into account when analyzing the state of technical security equipment and technical means of video surveillance in the event of similar situations in the future;

- analysis of pre-alarming, alarming and post-alarming video information coming from technical means of video surveillance;

- use of "sleep mode" to save power by the system;

- distribution and redistribution of computing resources between the central control center, security equipment and video surveillance equipment;

- the use of information about the distance to the place of violation, the parameters of the object of violation, the speed and direction of its movement through the guarded border to identify the object of violation by class (person, small animal or bird, vehicle), which gives additional information to the security service to detain the violator;

- the use of processing algorithms with a combination of signals from alarm sensors working on different physical principles;

- the use of neural network algorithms and fuzzy logic algorithms for intelligent information processing in a central control center,

many of which are not obvious and, as in the previous patent, may constitute objects of individual inventions, are not given in the description of the invention to the patent. We note that some of the indicated possibilities (for example, the use of the "sleeping" mode) have no relation to any kind of intellect (both artificial and natural). Reference to the fame and use of the neural network algorithm in security systems with an indication of the article by A.Yu. Zenova and N.V. Myasnikova (“Application of neural network algorithms in perimeter security systems” / News of higher educational institutions. Volga region. Technical sciences. - 2012. -№3 (23). - P. 15-24) is, in our opinion, insufficient justification for the feasibility of the declared opportunities, because the schemes and recommendations given in the article are too general.

Nevertheless, the presence of video surveillance in the "Intelligent Network Monitoring System of the Protected Area" allows with a greater probability than in the patent "Intelligent Security Alarm System with the ability to exchange information between detection means" that the cause of the operation of a sensor is a person, not animals, birds or other disturbing factors. This increases the reliability of the classification of objects of intrusion into the protected area and the validity of the decision to counteract the intruder (including due to the possibility of visual support of the intruder when moving around the protected area). The purpose of the invention is "providing the ability to monitor protected areas, taking into account the visual identification of the person-offender to distinguish him from animals, birds and other interference factors" - is achieved.

The disadvantage of this system is the complete dependence of the efficiency of its functioning on the human operator, the dependence on the "human factor", which in some cases leads to a decrease in the probability of detection, since it is the operator who, through monitoring the signals of the technical means of detection and video surveillance, visually classifies the triggered detection means objects, separating the intruder from various "interference factors" that trigger the detection means. Despite the highest recognition capabilities currently possible that no artificial intelligence can compare with yet, a number of significant drawbacks are inherent in the human operator. First of all, this is the limited ability to simultaneously monitor several objects, which in the case of group invasion of violators in different places on a protected area or with a large protected area requires several operators with the necessary experience not only individual but also team work. The disadvantages of having an operator must include other known negative manifestations of the “human factor”: exposure to fatigue and, as a consequence, the need for shift work, the possibility of distraction and, as a result, missing signals, the dependence of the degree of responsibility on individual character traits, the dependence of the quality of work is not only from training and work experience, but also from physical and emotional states (the negative impact of physical ailments, personal and family problems, ...), finally Alice probability of betrayal and deliberate sabotage. In addition, the human operator cannot be used in the protection of objects that pose a direct or potential threat to human health and life (for example, hazardous industries, areas with dangerous climatic conditions). To change operators, an appropriately equipped room with the necessary living conditions is required. Operators need to be trained (individual, and, if necessary, teamwork), to control their physical and moral readiness to take up duty, to control the quality of their service, they need to pay wages, pay the vacation period, finding them a replacement for this time, carry out their certification, periodically improve their qualifications, etc., which requires large time and financial costs.

In addition (including due to the limited capabilities of the human operator), technical surveillance systems in a particular surveillance area in the patent in question are turned on (in automatic mode or at the command of the operator) only when the technical means for detecting this area are triggered. Obviously, this method of engaging video surveillance systems, along with justified “saving of forces” (operator’s attention and power supply), reduces the likelihood of identifying an intruder in case of failure for one reason or another (including due to the targeted actions of the intruder) of the detection means. The reason for this failure may be the change in weather conditions mentioned in the patent, but the weather station used in the previous patent, which is necessary for this case, is not included in the description of the invention to the patent (in the description of the invention to the patent it is only indicated that “weather conditions are provided by obtaining the necessary information from an external weather station ”). The reason for missing a signal (failure of detection means in the presence of an intruder) may also be the presence of artificial anthropogenic interference of natural and artificial origin (operation of nearby mechanisms and machines, for example, vibration and acoustic noise from passing vehicles and flying aircraft, smoke from fires and industrial enterprises, electromagnetic radiation in various ranges from stationary and mobile objects (including flashes of light of various origins in the evening and at night, etc.). These interference, along with the parameters, are not only weather conditions (but generally natural, including, for example, earthquakes in active areas) it is also desirable to measure for accounting (including statistical) in information processing algorithms (knowledge of the statistics of their occurrence will make it possible to better determine artificial interference caused by the intruder or his escort group located outside the protected area). P The patent does not provide such measurements.

The closest known technical solution to the claimed one is “Intelligent intruder detection system” (patent RU 2665264 according to application No. 2016150799 with priority dated 12/22/2016), which contains security alarm sensors and an associated logical signal processing unit, a multilayer neural network and technical display facilities visual and audio information of the operator (the person making the decision), the outputs of the sensitive elements of the sensors being connected to the information inputs of the multilayer neural network, the output of the logical signal processing unit is connected to the information input "Alarm" of technical means for displaying visual and sound information for presentation to the operator (the person taking solution), the outputs of the multilayer neural network are connected to the information inputs of the technical means of displaying visual and audio information “Intruder”, “False alarm” and “Fault” for presentation to the operator (decision maker), and the outputs of the technical media The displays of visual and audio information in the form of a multi-vector information signal issued at the request of the operator are connected to the control inputs of a multilayer neural network. The operator (the decision maker), analyzing the visual and sound information displayed using the technical means of display, makes decisions on countering the intruder.

Signs common with the claimed device are: burglar alarm sensors, sensors associated with the sensors, a logical signal processing unit and an artificial neural network with training signals supplied to it.

The technical result of the invention, as indicated in the description, consists in "increasing the efficiency of the detection system of hostile and / or non-hostile intruders of protected stationary objects of special importance", and the presence of a neural network that gives signals "Intruder", "False alarm" and "Fault" to information the inputs of the technical means of displaying visual and sound information, according to the authors, "provides an increase in the reliability of assessing the nature of the impact of violators and the speed of making an adequate decision aimed at countering the intentions of violators."

Obviously, the presence of a neural network that analyzes the signals of the triggered sensors of the security alarm and makes a reliable conclusion about what caused the alarm - the actions of the intruder (in this case, the neural network gives a signal to the information input “Intruder” of the technical means for displaying visual and sound information) or the impact interference (a signal to the information input "False alarm"), and also detects a malfunction of the sensor or the information transmission circuit (signal to the information input "Fault") - and by transmitting this output to the operator (the decision maker), it allows to reduce the number of errors of the first kind ( taking sensors for triggering sensors from interference factors), as well as detecting sensor malfunctions in a timely manner. This ensures the achievement of the purpose of the invention is to increase the efficiency of detection of violators. And, in addition, partially freeing the operator from performing registration functions and providing him with information on the causes of the triggering and the state of the sensors that has been initially processed by the neural network, it really increases the “efficiency of the operator making an adequate decision” noted in the description of the invention.

However, one should point out some incorrectness of the statements used in the description of the invention, in particular, about increasing the “reliability of assessing the nature of the impact of violators” and “countering the intentions of violators using neural network technologies”. Given the capabilities and operation algorithms of modern artificial neural networks and analyzing the elements presented in the description of the invention, the connections between them and the operation of the claimed system, it is impossible to understand how to use the proposed artificial neural network to increase the "reliability of assessing the nature of the effects of violators" (based on the context of the description of the invention, Under the "nature of the influence of violators" is understood, probably, his "hostility" or "non-hostility") and understand their intentions? Of course, neither today, nor in the foreseeable future, it is impossible to distinguish a “hostile intruder” from a “non-hostile” one, and even more so to help understand its “purpose” using artificial neural networks that receive signals from traditional meters of physical quantities (sensors) (it should also be noted the absence in the description of the invention between the sensors and the neural network of the necessary transducers of the output signals of the sensors (analog or specific digital signal) to a normalized digital form (for example, in the range from 0 to 1) for input to d neural network). Moreover, the outputs of an artificial neural network, judging by the description of the invention and the logic of operation of modern artificial neural networks, are either “1” or “0”, fed to the information inputs of technical means of displaying visual and audio information: “Intruder”, “False” anxiety ”,“ Fault ”. The artificial neural network provided in the description of the invention does not provide any additional information. Therefore, there is no reason to really expand the effect of reducing the probability of false alarm through the use of a neural network to obtain the possibility of identifying the "hostility" or "non-hostility" of the offender, and even more so of his "design". The only way to form with some probability the assumption of “hostility” or “non-hostility” of a person identified in the protected area and with even less probability the assumption of his possible “design” consists in video surveillance of him for some time, which an experienced person should use with video cameras -operator of the security system, who also has the necessary psychological and other knowledge about the behavior of people in general and violators, in particular. But neither in the claims, nor in the description of the invention, the means of video surveillance are not mentioned and are not used.

The tacit assignment of a video camera to sensors is incorrect (the video camera may be connected to sensors or even include one or another sensor, for example, a motion sensor, but the sensor itself is not). Mention of the use of “simulation holograms” related to video surveillance (the possibility of targeted creation of which for video surveillance systems is, according to the authors, a disadvantage of the prototype they choose and the refusal to use video surveillance tools in the claimed system) is given in the description with reference to the formation of a training sample for a neural network, which improves the quality of recognition of the "image of the impact of trained, hostile violators" (it is difficult to understand what the "image of the impact" is). But even if there are video surveillance tools in the sensor unit, it is difficult to understand how the video information received from them is used, since their video signals are not sent to the operator, but to the logical processing unit, which, as follows from the description, “produces only the Alarm signal when there is offender ”(most likely, using the decisive rules mentioned in the previous patent known to date:“ two of three of them, ”etc.). In the best case, the Alarm signal can be supplemented by indicating the location (coordinates) and the type of sensor (s) that have triggered, which are displayed in the form with the help of technical means for displaying visual and sound information. It is also unclear how the use of the "multi-vector control signal" is technically carried out, as a result of which "the effectiveness of the proposed intelligent intruder detection system is increased when a training sample of interference exposure is generated by using the" Intruder "output signal (in essence," 1 "or" 0 " ) multilayer neural network? What is this “multi-vector signal”, why and how is it formed by “technical means of displaying visual and audio information of the operator” (judging by the given structural diagram of the invention)? It is unclear the absence of the operator mentioned in the description and claims of the invention described in the description of the invention, so that, judging by the scheme, “the decision to counteract the intruder” is taken by “technical means of displaying the operator’s visual and audio information”. Finally, the validity of the formula given in the description of the invention seems to be very weakly substantiated, which, in the authors' opinion, confirms the “positive effect of using the invention” (“increases by at least 10 ... 15% (?) The reliability of assessing the nature of the impact of violators and the speed of making an adequate decision aimed at countering violators using neural network technologies ”). Nevertheless, the use of the proposed artificial neural network, supplementing the traditional method of logical processing of signals from sensors (for example, according to the “two out of three” principle), allows to increase the probability of detecting intruders (both “hostile” and “non-hostile”) by reducing the number of false alarms and define the claimed system as intellectual (firstly, because its structure includes a representative of natural intelligence - the human operator, and secondly, following the current trend, refer to “Intellectual” everything that contains an artificial neural network). Therefore, this invention, despite the above highlighted "inaccuracies", is selected as a prototype.

The obvious disadvantages of the prototype include the following:

- insufficiently high probability of detecting unauthorized penetration (attempts of such penetration) into the protected area of a human intruder even if there is a professional human operator in the detection system, which is especially important for the protection of particularly important objects mentioned in the description of the invention. The reasons for this are as follows. Firstly, the use of only one (and only three-layer) artificial neural network of direct distribution, intended for independent decision-making about the cause of the sensors: “Intruder”, “False alarm”, “Fault”. Obviously, for the recognition of these situations, the training of the neural network should be carried out on a large number of examples, including: signals from triggered sensors due to the actions of the intruder, and these signals should be precisely signals (a function of time), because using threshold sensors alone is not enough); signals from triggered sensors due to the influence of other factors on them (animal, wind, device or technical complex with artificial intelligence (both remotely controlled and autonomous), anthropogenic sources: vibration from nearby cars, electromagnetic pickups of different nature, etc.) also in the form of time functions that differ from the triggering signals from the intruder; signals from sensors indicating their malfunction (the set is very large, because the cause can be either a primitive break (connection), or significantly changing the recorded signal and, as a result, the decision made about the cause of the operation, changing parameters and characteristics, in particular, multiplicative scale factor departure). To obtain a sufficient number of all these signals and create a representative training sample from them is very problematic. In addition, we note that this neural network can no longer be taught anything else (retrain in addition to what has already been learned), adding, for example, to the three available outputs (“Intruder”, “False alarm”, “Fault”) yet one or more other exits or dividing any existing exit into several (for example, dividing the Intruder output into two exits: “Hostile intruder” and “Non-hostile intruder” (following the terminology accepted in the description of the invention). Everything learned when a new exit appears will be forgotten ("catastrophic forgetfulness" is a significant drawback of modern artificial neural networks). Secondly, the lack of video surveillance. Thus, the operator does not have the opportunity to monitor the protected area and the events taking place on it. Since a person receives at least 80% of information about the outside world through vision, then the triggering signals of sensors of physical quantities without the use of video surveillance, even backed up by information from artificial a neural network is clearly not enough for the operator to confidently recognize the intruder (both "hostile" and "non-hostile");

- the inability to identify autonomous or controlled (including remotely) technical (robotic) means and complexes, as well as trained animals and birds, as unauthorized persons who have penetrated (trying to penetrate) into the protected territory by land or air; The triggering of sensors from such intruders in the prototype will be referred to as “False alarm” or “Fault”. But it is precisely such violations, first of all, with the use of robotic (including miniature) tools and systems that are already being successfully implemented, and in the coming years will become the main and most dangerous. The effectiveness of measures taken to counteract very significantly depends on who the offender is: a person, a robot (ground-air, large-small), animal, bird, ...;

- the impossibility of identifying the offender. Meanwhile, recognition in an intruder, for example, of a specific person, allows obtaining additional information about him (motives, psyche, acquaintances, physical abilities, ...) from a centralized database, which is also very important for taking adequate measures;

- the impossibility of the detection system without a human operator. The mandatory presence of a human operator, firstly, limits the scope of the security system (the system cannot be used when guarding objects that pose a direct or potential threat to human health and life, which include, for example, hazardous industries, landfills for various purposes, etc.) . Secondly, it significantly increases the cost of protection, since an operator needs an appropriately equipped room with the necessary conditions for life support, the operator must be trained (individual, and, if necessary, and team work), control physical and moral readiness to take up duty, control the quality of service, he needs to pay wages, pay the vacation period, finding him a replacement for this time, conduct certification, periodically improve his qualifications. It should be noted that with a large protected area (a large number of objects in the protected area), simultaneous penetration (in different places) of several violators, several operators with high coordination of team work may be required, which requires additional costs (for training, etc.). Thirdly, it limits the security system's ability to identify specific objects (for example, people), since the personal database of a human operator is small compared to modern centralized computer databases used in intelligent systems. Fourth, it creates certain risks of failure to fulfill the protection task due to a significant impact on the protection of the “human factor” system: the ability to be distracted (both consciously and unconsciously), exposure to fatigue (which requires shift work, which additionally increases the cost of protection), dependencies responsibility from individual character traits, physical and emotional states, and finally, the possibility of betrayal, targeted wrecking.

SUMMARY OF THE INVENTION

The technical result (goal) of the invention is to increase the likelihood of detecting unauthorized entry (attempts of unauthorized entry) into a protected area and the movement of natural violators (human, trained animal or bird) or artificial (controlled, including remotely, or autonomous ground or an airborne vehicle) of origin capable of creating (creating) a threat to the protected object by assigning the identified violator to a predetermined class of violators, as well as obtaining the possibility (if there is sufficient a priori information) of identifying the violator within this class and the possibility of reducing the cost of operating the system and its use in conditions incompatible with a long stay of a person (hazardous production, underwater and (or) space objects, etc.) due to the optional presence of a human operator. At the same time, the claimed intelligent automatic intruder detection system (IASON) contains many artificial neural networks. Despite the fact that a person has the highest integrated recognition capabilities that no artificial intelligence can compare with yet, in certain narrow areas of recognition, artificial intelligence, implemented on the basis of artificial neural networks of the so-called “deep learning”, has already surpassed humans (games, medical diagnoses, face recognition, ...), artificial neural networks are capable of high-quality processing of signals from detection tools, the number of types and capabilities of which significantly exceed the number and capabilities of the human senses. A variety of detection tools and capabilities of modern highly specialized artificial neural networks according to classification and identification are used in the claimed invention.

The technical result is achieved by the fact that the claimed IASON, in addition to means for detecting intruders in the form of burglar alarm sensors operating on various physical principles of operation associated with sensors of a logical signal processing unit and an artificial neural network with training signals supplied to it, contains:

- means of detecting intruders in the form of imaging tools for violators in various frequency ranges (video cameras, thermal imagers, radar, etc.), which together with contact and distant burglar alarm sensors operating on various physical principles of operation (radio wave, radio beam, infrared, vibration ( vibrometric), seismic, acoustic, magnetometric, capacitive, inductive, chemical, radiation, etc.) are combined into N blocks of detection tools, each of which contains detection devices of the same type (sensors operating on the same physical principle of operation, or image forming means in the same frequency range);

- devices for recording and primary processing of signals from detection means, combined into N blocks of registration and primary processing of signals in such a way that all devices of each block can receive and process signals from detection means of only one type;

- artificial neural networks that recognize signals from detection means and are combined into N blocks of artificial neural networks in such a way that all neural networks of each block are able to recognize signals from detection means of only one type;

- a block of means for recording the parameters of natural factors (meteorological data using a digital weather station, if necessary, seismic data using a seismic station, data on the magnitude and disturbances of the Earth’s magnetic field using magnetometers) and a block of means for recording the parameters of anthropogenic factors (vibration and noise from stationary and mobile equipment, electromagnetic radiation in various ranges, dust formations from industrial enterprises, etc.) that can affect the operation of detection tools and other elements of IASON;

- a central control center comprising a logical signal processing unit, a data flow control unit, a memory unit, an environmental parameter accounting unit, a final recognition unit and a decision making unit,

moreover:

- blocks of detection tools, blocks of devices for recording and primary signal processing and blocks of artificial neural networks are combined in N detection channels so that each detection channel contains one block of detection tools, one block of devices for recording and primary signal processing and one block of artificial neural networks;

- in each detection channel, the output of the block of detection means is connected to the input of the block of registration devices and primary signal processing, the main output of which is connected to the main input of the block of artificial neural networks, the additional outputs of all blocks of devices of registration and primary signal processing are connected to the input of the block of logical signal processing, the output of which is connected to additional inputs of the data flow control unit and the final recognition unit, the outputs of all artificial neural network blocks are connected to the main input of the data flow control unit, the outputs of which are connected to the main inputs of the final recognition unit, the outputs of the block of means for recording the parameters of natural factors and the registration unit parameters of anthropogenic factors are associated with the input of the environmental parameters accounting unit, the outputs of which are connected to the control inputs of all the detection means blocks and the additional input of the final recognition unit, the main output of which is connected with the main input of the decision block associated with direct and feedback connections with the memory block, which is also connected with direct and feedback connections with the environmental parameters accounting block and the final recognition block, and the additional output of the final recognition block is connected with the training inputs of all blocks of artificial neural networks;

- the inputs and outputs of the blocks of detection means, blocks of registration devices and primary signal processing, blocks of artificial neural networks are represented by the inputs and outputs of detection means, registration devices and primary signal processing, artificial neural networks, respectively;

- each n-th block of detection tools included in the n-th detection channel, n = [l, N], contains K _n sensor areas located in different known places of the protected area, each k-th, k = [1, K _n ] the sensor area consists of M _{n, k of the} same type of detection means, each means of detection D _{n, k, m} , m = [1, M _{n, k} ] of the sensor area is characterized by known location coordinates in the protected area and (if necessary) individual settings ;

- each n-th detection channel and the primary signal processing unit included in the n-th detection channel contains K _n devices for recording and primary signal processing P _{k, n} , the input of each k-th, k = [1, K _n ] of which is associated with all outputs of the detection means of the k-th, k = [1, K _n ] sensor area of the detection means;

- each n-th block of artificial neural networks included in the n-th detection channel contains K _n artificial neural networks C _{n, k} , the input of each k-th, k = [1, K _n ] of which is connected with the main output of the k-th , k = [1, K _n ] of the device for registration and primary signal processing;

- the number of outputs of each artificial neural network C _{n, k of} each block of artificial neural networks is the same and equal to L - the number of classes of detected intruders.

At the same time, the outputs of the IASON are the outputs of the decision-making unit, which are formed on which the real-time results of the work of the IASON provide information about violations and violators, about failures, sufficient for the prompt adoption of adequate measures by the quick response team to counter violators, the repair team to eliminate the malfunction, fire team to extinguish the fire, etc., and also carry information about the system for recording it on an external data storage server.

BRIEF DESCRIPTION OF THE DRAWINGS

Schemes of the claimed IASON are shown in FIG. 1 and FIG. 2. In FIG. 1 marked: 1, 2, 3, 4 - channels for detecting violators (circled by a dotted line: first, second, n-th, N-th, respectively); 5, 6, 7, 8 - blocks of detection means (BSO): BSO 1, BSO 2, BSO n, BSO N, respectively; 9, 10, 11, 12 - blocks of devices for recording and primary signal processing (BURPOS): BURPOS 1, BURPOS 2, BURPOS n, BURPOS N, respectively; 13, 14, 15, 16 - blocks of artificial neural networks (SINS): SINS 1, SINS 2, SINS n, SINS N, respectively; 17 - logical signal processing unit (BLOS), 18 - data flow control unit (BAPD), 19 - total recognition unit (BIR), 20 - decision making unit (BPR), 21 - memory unit (BP), 22 - means block registration of parameters of natural factors (BSRPPF), 23 - a block of means for recording parameters of anthropogenic factors (BSRPAF), 24 - a unit for recording environmental parameters (BUPVS), 25 - a central control point (circled by a dotted line), 26 - quick response team (KBR), 27 - repair team (RB), 28 - data storage server (SHD). In FIG. 2 is a diagram of each n-th, n = [l, N] detection channel, on which are indicated: 29, 30, 31 — means for detecting the first sensor region of the nth block of detection means 7; 32, 33, 34 - detection means of the second touch region of the nth block of detection means 7; 35, 36, 37 - means for detecting the k-th sensor region of the nth block of detection means 7; 38, 39, 40 - detection means K of the _n- th sensor region of the nth block of detection means 7; 41, 42, 43, 44 - registration devices and primary signal processing (URPOS) of the n-th block of devices for recording and primary signal processing 11 from the first, second, k-th and K _n- th sensor areas of the block of detection means 7, respectively; 45, 46, 47, 48 - an artificial neural network (ANN) n-th block of artificial neural networks 15 which recognize signals from the first, the second, k-th and K _n -th detection means sensory areas unit 7 respectively fed to them via device registration and primary signal processing 41, 42, 43, 44 of the block device registration and primary signal processing 11, respectively.

DETAILED DESCRIPTION OF THE INVENTION

IASON works as follows. Since all detection channels work the same way, consider the operation of the nth detection channel 3. When an intruder enters a protected area (when attempting such an entry) and moves along it (or at an unacceptable height above it), this or that detection means D _{n k, m} , m = [l, M _{n, k} ] included in the k-th (k = [1, K _n ]) sensory region of the nth (n = [l, N]) block of detection tools. For example, assume that the detection means D _{n, k, l} 35, which enters the sensor region containing M _{n, k} of the same type of detection means, has worked. The signal from this detection means is supplied to the corresponding device for recording and primary signal processing P _{n, k} 43, which, firstly, registers the fact, coordinates and response time with the transmission of this information to the signal processing unit 17 of the central control center 25, and, secondly, it converts the signal received from the triggered detection means D _{n, k, 1} 35 into a normalized digital set, which is then fed to recognize the source that caused it to the input of the corresponding artificial neural network C _{n, k} 47. and each artificial neural network 45, 46, 47, 48 of the block of artificial neural networks 15, at the stage of preliminary (previous to the operation) training, is configured to recognize intruders of all classes to be recognized based on the analysis of signals from detection devices of the same type. At the output of the artificial neural network 47, signals arriving through the data flow control unit 18 to the final recognition unit 19 are formed, showing to which 1st class of the possible, ι = [1, L], the intruder belongs. Signal values can be either “1” at one output and “0” at the others, or the probabilities of the intruder belonging to each of the classes if unity is the sum of all probabilities. In the conditions of intersection of class features and the presence of interference of natural and artificial origin, probabilistic separability of classes is preferable (currently probabilistic neural networks are widely used). The data flow control unit 18, taking into account the information from the logical signal processing unit 17, generates a sequence (sequence) of supplying to the input of the final recognition unit 19 signals from the outputs of artificial neural networks 45 ... 48 of the block of artificial neural networks 15 in case of their simultaneous (in one clock cycle) receipts.

Parameters of natural factors (meteorological data, data on seismic activity, parameters of the Earth’s magnetic field and its perturbations) and parameters of anthropogenic factors (vibration and noise from equipment, electromagnetic interference, dust formations, etc.) that can affect the operation of intruder detection tools 29 ... 40 and other elements of IASON, are measured using the block of means for recording the parameters of natural factors 22 and the block of means for recording the parameters of anthropogenic factors 23 and fed to the block for recording environmental parameters 24. The final recognition block 19, taking into account information from the logical signal processing block 17 about the place and the response time of detection tools 29 ... 40, information on environmental parameters (natural: temperature, pressure, humidity, wind speed, precipitation parameters (rain, snow, hail), etc. and man-made: parameters of electromagnetic radiation in various ranges, vibrations, smoke, dust, noise, etc.) at the time of operation detection 29 ... 40 (this allows you to determine the degree of their influence on the accuracy and reliability of the signals of the detection means) from the unit for recording environmental parameters 24 and the information stored in the memory unit 21 (database, knowledge base in the form of species (phylogenetic) and individual (ontogenetic) operating experience) processes (in the general case, spatio-temporal) recognition signals from artificial neural networks 45 ... 48 and, firstly, makes the final decision on the class of the intruder, and secondly, if necessary and there is sufficient a priori information in the database this class with the help of artificial neural networks included in its composition (at least L networks are required to identify violators of all classes) identifies the violator by its individual characteristics and, thirdly, recognizes situations associated with malfunction of detection tools and (or) transmission channels information, the presence of abnormal situations of operation, issuing in decision block 20 all this information, and into artificial neural networks 45 ... 48 of the corresponding detection channels 1 ... 4 - training signals that allow neural networks 45 ... 48 to replenish their individual experience. In this case, the appearance of the second and subsequent triggering of detection tools from the actions of the intruder narrows the domain of definition (reduces the number) of classes to which the intruder can be attributed with a non-zero probability. The accumulation and processing of information about the facts of detection means operation during the penetration of single and group violators into the protected area and movement along it, about the spatio-temporal parameters of penetration and movement, about the decisions made by IASON and their results allows neural networks 45 ... 48 and neural networks used for identification networks of the block of final recognition replenish (including - in real time) their individual experience with improving the quality of subsequent detection processes.

All information is stored in the memory block 21. The decision block 20, depending on the results of recognition and identification in real time, notifies (with the provision of the full necessary information) a quick response command 26 (it can be located outside the central control center and even outside the protected area) and (or) repair team 27 (as well as a special team, such as a fire brigade). In addition, through the decision block, the current information on the operation of IASON is transmitted to an external data storage server 28.

The number of L classes of violators and their signs are set at the IASON development stage and depend on the type of protected object, the size of the protected area, natural conditions, interference anthropogenic factors, requirements to the quality indicators of IASON functioning. Moreover, the features of each class used in recognition may partially overlap. As an extended version, the following types of classes can be distinguished: man, large animal, small animal, bird, anthropomorphic robot, large technical means moving on the ground, small technical means moving on the ground, large technical means moving on the air, small technical means moving on the air . Recorded signs can be as static: overall dimensions (height, width, length), weight, body shape of the intruder and his elements, postures, gestures, temperature, images (“photographs”) in various frequency (spectral) ranges, chemical composition, presence ferromagnetic materials, etc., as well as dynamic: speed and acceleration of movement, vibrations of the body and its parts during movement, emitted sounds, signals during shock loads on the ground or the fence during movement (climbing), etc. When triggered by two or more detection tools in the same sensor area or different sensor areas, the registration and primary processing device registers the facts of the operation, transmitting information about the facts of the operation and the corresponding time points to the logic processing unit 17 of the central control point 25, normalizes the received operation signals and sequentially ( the sequence is determined either by the moments of arrival of the signals, or by the importance of the sensory region from which they came) sends them to the corresponding neuron th network.

устройств регистрации и первичной обработки -

искусственных нейронных сетей в каналах обнаружения -

При этом конкретные места размещения средств обнаружения того или иного типа, их сочетание со средствами обнаружения других типов определяются их информативностью (например, видеокамера - высокоинформативное, а датчик обрывного типа -малоинформативное средство), взаимодополнительностью (возможностью получения синергетического эффекта при их объединении), надежностью, задаваемыми требованиями, условиями функционирования и пр. В целом при размещении средств обнаружения на конкретном участке охраняемой территории целесообразно их сочетать таким образом, чтобы регистрируемые от них сигналы несли информацию о разных параметрах нарушителей, что дает желаемый синергетический эффект. При этом средства обнаружения могут быть как стационарными, так и мобильными (например, с заранее заданным маршрутом патрулирования), использующими для передачи и получения информации как проводную связь, так и радиосвязь в различных частотных диапазонах, а также их комбинации. В случае использования радиосвязи возможна самоорганизация применительно как к формированию маршрутов передачи информации (например, по критерию наилучшего качества), так и к размещению (мобильных) средств обнаружения на территории по известным алгоритмам работы самоорганизующихся сенсорных сетей. Знание моментов времени и координат срабатывания разных средств обнаружения позволяет принять обоснованные решения о наличии одного или нескольких нарушителей. Информативной является и траектория движения. Учет временного фактора и последовательности действий нарушителя, уязвимости и важности объектов на охраняемой территории, имеющегося опыта эксплуатации данной и других ИАСОН позволяет осуществлять достоверное прогнозирование поведения нарушителя, оперативно принимать адекватные (в том числе - предупреждающие) действия.The total number of detection tools is

registration and preprocessing devices -

artificial neural networks in the detection channels -

At the same time, specific locations for detecting this or that type, their combination with other types of detecting means are determined by their information content (for example, a video camera is highly informative, and a discontinuous type sensor is not very informative), complementarity (the possibility of obtaining a synergistic effect when combined), reliability defined by requirements, operating conditions, etc. In general, when locating detection equipment in a specific area of the protected area, it is advisable to combine thief so that signals recorded on them carry information about different parameters of violators that gives the desired synergistic effect. In this case, the detection tools can be both stationary and mobile (for example, with a predetermined patrol route), using both wire communication and radio communication in various frequency ranges, as well as their combinations, to transmit and receive information. In the case of the use of radio communications, self-organization is possible in relation to both the formation of information transmission routes (for example, according to the criterion of the best quality) and the placement of (mobile) detection tools in the territory according to well-known algorithms of self-organizing sensor networks. Knowing the times and coordinates of the operation of different detection tools allows you to make informed decisions about the presence of one or more intruders. The trajectory of movement is also informative. Taking into account the time factor and the sequence of actions of the intruder, the vulnerability and importance of objects in the protected area, the existing experience in operating this and other IASONs, it is possible to carry out reliable prediction of the behavior of the intruder and to take adequate (including warning) actions quickly.

It should be noted that the species (phylogenetic) experience laid down in IASON during its manufacture reflects the experience of designers in creating both security systems in general and the experience of applying and improving IASON of a specific type in the protection of specific typical objects in specific typical conditions. IASON individual experience is formed on the basis of "genetic programs" in the process of primary education of IASON and is further improved and replenished in the process of its operation in real conditions through, first of all, retraining (by supplying training signals) of artificial neural networks of detection channels 1 ... 4 and (with necessity) of the block of final recognition 19 based on the current results of their work, as well as in the process of planned periodic training of IASON in "refresher courses".

Claims (13)

1. Intelligent automatic system for detecting intruders, containing detection tools in the form of security sensors, logical signal processing unit and artificial neural network with training signals supplied to it, characterized by the presence of additionally introduced detection tools in the form of imaging tools for violators, combined together with burglar alarm sensors in N blocks of detection tools, N blocks of registration devices and primary processing of signals from detection tools, N blocks of artificial neural networks that recognize signals from detection tools, data flow control unit, block of means for recording parameters of natural factors, block of registration tools parameters of anthropogenic factors, a unit for accounting for environmental parameters, a memory unit, a final recognition unit and a decision making unit, the blocks of detection means, the blocks of registration devices and the primary image signal bots and blocks of artificial neural networks are combined into N detection channels in such a way that each detection channel contains one block of detection tools, one block of recording and primary signal processing devices and one block of artificial neural networks, while in each detection channel the output of the block of detection tools connected to the input of the block of devices for recording and primary signal processing, the main output of which is connected to the main input of the block of artificial neural networks, the additional outputs of all blocks of devices for recording and primary signal processing are connected to the input of the block of logical signal processing, the output of which is connected to additional inputs of the flow control unit data and the final recognition unit, the outputs of all blocks of artificial neural networks are connected to the main input of the data flow control unit, the outputs of which are connected to the main inputs of the final recognition unit, the outputs of the unit for recording parameters of natural factors and the unit for recording the parameters of anthropogenic factors are associated with the input of the environmental parameters accounting unit, the outputs of which are connected to the control inputs of all blocks of the detection tools and the additional input of the final recognition unit, the main output of which is connected to the main input of the decision unit associated with direct and feedback connections a memory block, which is also connected by direct and feedback connections to the environmental parameters accounting unit and the final recognition unit, and the additional output of the final recognition unit is connected to the training inputs of all blocks of artificial neural networks. 2. The system according to p. 1, characterized in that various types of burglar alarm sensors are used as detection means that implement various physical principles of operation: radio wave, radio beam, infrared, vibration (vibrometric), seismic, acoustic, magnetometric, capacitive, inductive, chemical, radiation. 3. The system according to claim 1, characterized in that various types of imaging means of intruders operating in different frequency ranges are used as detection means: optical (video cameras), infrared (thermal imagers), radio range (radars). 4. The system according to claim 1, characterized in that each block of detection means contains detection means of only one type (sensors operating on the same physical principle of operation, or means of imaging intruders operating in the same frequency range). 5. The system according to claim 1, characterized in that the recording and primary signal processing devices of each block of registration and primary signal processing devices are pre-configured to receive and process signals from detection devices of only one type. 6. The system according to claim 1, characterized in that the artificial neural networks of each block of artificial neural networks are pre-trained in recognizing signals from detection devices of only one type. 7. The system according to claim 1, characterized in that the inputs and outputs of the blocks of detection means, blocks of registration devices and primary signal processing, blocks of artificial neural networks are represented by the inputs and outputs of detection means, registration devices and primary signal processing, artificial neural networks, respectively. 8. The system according to claim 1, characterized in that each n-th block of detection means included in the n-th detection channel, n = [1, N], contains K _n sensory areas located in different known places of the protected area, each k-th, k = [1, K _n ] sensory area consists of M _{n, k of the} same type of detection means, each means of detection D _{n, k, m} , m = [1, M _{n, k} ] of the sensor region is characterized by known coordinates in a protected area and (if necessary) individual settings. 9. The system according to claim 1, characterized in that each n-th detection and primary signal processing unit included in the n-th detection channel contains K _n devices for recording and primary signal processing P _{n, k} , the input of each k-th, k = [1, K _n ] of which is associated with all outputs of the detection means of the k-th, k = [1, K _n ] of the sensor area of the detection means. 10. The system according to claim 1, characterized in that each n-th block of artificial neural networks included in the n-th detection channel contains K _n artificial neural networks C _{n, k} , the input of each k-th, k = [1, K _n ] of which is connected with the main output of the k-th, k = [1, K _n ] device for recording and primary signal processing. 11. The system according to claim 1, characterized in that the number of outputs of each artificial neural network C _{n, k of} each block of artificial neural networks is the same and equal to L - the number of classes of detected intruders. 12. The system of claim 1, wherein the logical signal processing unit, the data flow control unit, the memory unit, the final recognition unit, the environmental parameter accounting unit, and the decision making unit form a central control point. 13. The system according to p. 1, characterized in that its outputs are the outputs of the final decision block, which carry information about the revealed violations and violators, as well as the operation of the system and its malfunctions.

Images