CN111814681A

CN111814681A - Method and device for automatically identifying instability risk megaevent of construction worker based on sensor

Info

Publication number: CN111814681A
Application number: CN202010654857.0A
Authority: CN
Inventors: 郭红领; 林啸
Original assignee: Tsinghua University
Current assignee: Tsinghua University
Priority date: 2020-07-09
Filing date: 2020-07-09
Publication date: 2020-10-23

Abstract

The invention discloses a sensor-based automatic identification method and device for a construction worker instability megaevent, wherein the device comprises the following steps: a carrier body for carrying the identification device; the acquisition device is used for acquiring event identification data; an external prompter; and the controller is used for controlling the external prompter to send out a destabilization dangerous megaevent alarm when the occurrence of the destabilization dangerous megaevent is recognized according to the event recognition data. The device is based on monitoring human gesture stability, considers construction environment and operation mode to can be under the condition that does not influence workman's operation, the unstable motion of real-time effective monitoring takes place, and has high expansibility.

Description

Method and device for automatically identifying instability risk megaevent of construction worker based on sensor

Technical Field

The invention relates to the technical field of building construction safety, in particular to a sensor-based method and a sensor-based device for automatically identifying a construction worker instability risk megaevent.

Background

The building industry is one of the pillar industries for promoting national economic development and social progress, but the building industry is also the high-safety-accident issuing industry, and the safety problem is more and more serious. Since 2012, the number of people dead caused by safety accidents in the construction industry in China exceeds that in the coal mine industry, and the construction industry becomes the most dangerous industry in all industrial production fields in China. In the year 2018, the number of dead 840 people in China caused by 734 safety accidents of house municipal engineering production is increased by 6.1% and 4.1% compared with the number of dead 840 people in the last year. Among them, high fall, object strike, collapse, mechanical injury and lifting injury are the main five types in construction safety accidents.

Most workers experience instability (LOB), which is particularly large for workers working at high altitudes. Many researchers have determined that there is a strong correlation between a fall incident and instability, and poor body posture stability is one of the major factors contributing to increased risk of falls. A number of documents demonstrate that there is a close relationship between a fall incident and an LOB, and monitoring and analysing the balance of a worker in real time may help to identify the precursor to a fall, and thus prevent it.

On the other hand, instability is also a consequence of many unsafe behaviors, such as walking without paying attention to obstacles ahead, trying to lift heavy objects out of the strength range, getting too close to the edge, etc. The occurrence of the instability conditions is monitored and analyzed in real time, the behavior of workers is improved, and the construction safety is enhanced.

There have been some studies that successfully use inertial measurement units (e.g., accelerometers and gyroscopes) to distinguish between different types of motion. The inertial measurement unit has the advantages of low cost, portability, small volume and easiness in monitoring daily activities. The inertial measurement unit can measure velocity, acceleration, direction and gravity, and the acceleration data can be used to monitor physiological conditions of the human body, such as posture stability and the like. However, most of the methods do not focus on the construction field, which causes many of the methods to have various restrictions under construction environments.

Disclosure of Invention

The present invention is directed to solving, at least to some extent, one of the technical problems in the related art.

Therefore, the invention aims to provide a sensor-based automatic construction worker instability risk megaevent recognition device which can effectively monitor instability action occurrence in real time without influencing the operation of workers and has high expansibility.

The invention also aims to provide a sensor-based automatic construction worker instability risk megaevent identification method.

In order to achieve the above object, an embodiment of the invention provides a sensor-based automatic identification device for a megaevent of instability of a construction worker, including: a carrier body for carrying the identification device; the acquisition device is used for acquiring event identification data; an external prompter; and the controller is used for controlling the external prompter to send out a destabilization risk megaevent alarm when the occurrence of the destabilization risk megaevent is recognized according to the event recognition data.

The sensor-based automatic construction worker instability risk megaevent recognition device provided by the embodiment of the invention is based on the monitoring of the stability of the human body posture, takes the construction environment and the operation mode into consideration, can effectively monitor the occurrence of instability actions in real time under the condition of not influencing the operation of workers, and has high expansibility.

In addition, the sensor-based automatic identification device for the megaevent of the instability of the construction worker according to the above embodiment of the present invention may further have the following additional technical features:

further, in one embodiment of the invention, the event identification data includes one or more of three-axis acceleration, three-axis angle, three-axis magnetic data, and air pressure value.

Further, in an embodiment of the present invention, the method further includes: and the power supply equipment is used for supplying power to the identification device.

Further, in one embodiment of the invention, the external cue comprises at least one acoustic warning device and at least one optical warning device.

Further, in an embodiment of the present invention, the method further includes: and the communicator is used for sending the event information of the instability risk megaevent to a preset terminal.

Further, in an embodiment of the present invention, the method further includes: and the recording device is used for recording the field data of the constructors.

Further, in one embodiment of the present invention, the carrier body is on a wearable device.

Further, in one embodiment of the present invention, the wearable device includes a safety helmet, a belt, and glasses.

In order to achieve the above object, an embodiment of another aspect of the present invention provides a method for automatically identifying a megaevent of instability of a construction worker based on a sensor, including the following steps: collecting event identification data; and when the instability risk megaevent is identified according to the event identification data, sending out an instability risk megaevent alarm.

The method for automatically identifying the instability risk megaevent of the construction worker based on the sensor is based on the monitoring of the human body posture stability, takes the construction environment and the operation mode into consideration, can effectively monitor the occurrence of instability action in real time under the condition of not influencing the operation of the worker, and has high expansibility.

In addition, the automatic identification method for the instability risk megaevent of the construction worker based on the sensor according to the embodiment of the invention can also have the following additional technical characteristics:

Further, in an embodiment of the present invention, the method further includes: and sending the event information of the instability risk megaevent to a preset terminal.

Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.

Drawings

The foregoing and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:

fig. 1 is a schematic structural diagram of a sensor-based automatic identification device for a megaevent of instability of a construction worker according to an embodiment of the invention;

FIG. 2 is a design connection diagram and a physical assembly diagram of a sensor-based automatic identification device for the megaevent of instability of a construction worker according to an embodiment of the invention;

FIG. 3 is a flow chart of an algorithm according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of a relationship between a module placement direction and a measurement coordinate system according to an embodiment of the present invention;

FIG. 5 is a schematic view of a survey coordinate system and a geodetic horizontal coordinate system according to an embodiment of the invention;

FIG. 6 is a schematic diagram of a feature calculation method according to an embodiment of the present invention;

FIG. 7 is a schematic diagram of a decision tree according to an embodiment of the present invention;

FIG. 8 is a schematic diagram illustrating continuous event frame discrimination according to an embodiment of the present invention;

FIG. 9 is a schematic diagram of an experiment according to an embodiment of the present invention;

fig. 10 is a flowchart of a method for automatically identifying a destabilization risk megaevent of a construction worker based on a sensor according to an embodiment of the invention.

Detailed Description

Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention.

The method and the device for automatically identifying the megaevent of instability of a construction worker based on a sensor according to an embodiment of the present invention will be described below with reference to the accompanying drawings.

Fig. 1 is a schematic structural diagram of a sensor-based automatic identification device for a megaevent of instability of a construction worker according to an embodiment of the invention.

As shown in fig. 1, the sensor-based automatic identification device 10 for megaevent of instability of construction workers comprises: carrier body 100, collection device 200, external cue 300 and controller 400.

Wherein the carrier body 100 is used for carrying an identification device; the acquisition device 200 is used for acquiring event identification data; the controller 400 is used for controlling the external prompter to send out a destabilization dangerous megaevent alarm when the occurrence of the destabilization dangerous megaevent is recognized according to the event recognition data. The device 10 of the embodiment of the invention can effectively monitor the occurrence of instability action in real time under the condition of not influencing the operation of workers, and has high expansibility.

It is understood that the apparatus 10 of the embodiment of the present invention includes a data acquisition device, a controller, an external prompter, a power supply device and an apparatus carrier. When designing this device, attributes such as accuracy, real-time, convenience, economic nature have been considered, specifically: the accuracy is required, the data acquired by the device should be as accurate as possible, and meanwhile, the device has certain environmental interference resistance; the device can receive, process and feed back information in real time to achieve the effect of real-time reflection; convenience, the device should be as easy to wear as possible, and when worn, should not interfere with the normal operation of the worker; economy requires that the device be constructed using components that are easy to assemble and relatively low cost to increase the likelihood of its deployment. In view of the above requirements, the device 10 according to the embodiment of the present invention is shown in fig. 2.

In one embodiment of the invention, the carrier body 100 is on a wearable device, which includes a hard hat, a belt, and glasses.

It will be appreciated that the carrier's convenience considerations are primarily based on the wearability of the device. In construction, a safety helmet is a safety device that every worker must wear while working. The helmet is an ideal device carrier because of its good wearability, mandatory wearability, and fixity, and can fix the whole device at the head position of a person.

It is understood that the collecting device 200 may be a sensor, and those skilled in the art can select a specific sensor type according to actual situations, and is not limited to the specific type.

As a possible implementation, the sensor: JY901B can be used as a measurement index sensor. The JY901B sensor is a 10-axis sensor, is provided with a gyroscope, an accelerometer, a magnetic field and an air pressure sensor, can actively output three-axis acceleration, three-axis angles, three-axis magnetic data and air pressure values, and can carry out serial port communication with a GPS (global positioning System) chip commonly found on the market. Meanwhile, JY901B is provided with a dynamic Kalman filtering algorithm and a magnetic field compensation algorithm, and can output numerical values more accurately in a dynamic environment and a micro-magnetic environment. JY901B has higher anti-interference performance and lower cost on the premise of meeting the requirement of accurate data acquisition, and is used accordingly. JY901B sensor needs to be kept as horizontal as possible during installation.

Further, in one embodiment of the present invention, the external cue 300 comprises at least one acoustic warning device and at least one optical warning device.

It can be understood that the external prompter 300 is an external feedback device, the acoustic alarm device may be a buzzer, a voice module, etc., and the optical alarm device may be an LED (Light Emitting Diode) indicator Light, etc.

For example, an acoustic alarm device is a buzzer, an optical alarm device is an LED indicator light, for example, a red-green double-color LED indicator light is connected to a raspberry, and when the system identifies the instability action, the red light is turned on; the green light is turned on for the rest of the time. A buzzer is connected to the raspberry and buzzes when the system recognizes a destabilizing action. The external indicator can visually indicate whether the instability action is identified or not, and reminds workers wearing the device to pay attention.

Further, in an embodiment of the present invention, the controller 400 is a central processing unit, and a person skilled in the art may specifically select a type of the central processing unit according to an actual situation, which is not specifically limited herein. For example, the central processing unit may be a single-board computer with a linux system, such as raspberry pi (RasberryPi) Zero W, which can perform a lightweight computing task, has WIFI and bluetooth modules, has a size of 6.5cm by 3.0cm, and is very suitable for being applied to embedded devices. JY901B sensor will be in serial port communication with raspberry pi through I2C.

The preset terminal can be a computer or a mobile phone and the like, the communicator can be an information collection device, and particularly, the raspberry group can remotely communicate with a computer under the same local area network through an SSH (secure Shell protocol), so that the starting and closing of the control device are controlled, and the instability signal can also be transmitted to the computer or the mobile phone through wireless network communication to check and record data and remind a manager to pay further attention.

Further, in an embodiment of the present invention, the method further includes: a power supply device and a recording apparatus. The power supply equipment is used for supplying power to the controller, and the recording device is used for recording field data of constructors.

It is to be understood that the power supply device may be a battery module, a lithium battery module, or the like, and the recording apparatus may be a camera module. Taking the power supply device as a storage battery module and the recording device as a camera module as an example, the storage battery module supplies power to the controller, and the camera module is used for recording the camera module on the site.

Further, the sensor-based automatic identification device 10 for the megaevent of instability of a construction worker works as follows:

the JY901B module collects acceleration data and angle data in real time. Since the device is fixed to the head of the worker by means of a safety helmet, this means that the acceleration data and the angle data can be regarded as acceleration and angle data of the head movement of the worker. The data will be transmitted to the raspberry pi Zero, which will analyze the current data in real time according to a predetermined algorithm to determine whether the current data characteristics match the destabilized data characteristics. If the light source is matched with the raspberry pi, the LED lamp is controlled to turn red, and the buzzer buzzes for 5 seconds. Within 5 seconds, no new action is needed, and repeated alarming is prevented.

On the other hand, the raspberry group sends information such as the current time, the worker number and the like to a control center through a wireless network by using short messages, mails or other modes, and the control center can be a computer of a manager or a smart phone. If the function of the device is further expanded, the position of a worker, a real-time picture and the like can be sent. The administrator can check the potential safety hazard, change the work arrangement and the like according to the received data.

Further, the algorithm design of the sensor-based automatic identification device 10 for the megaevent of instability of construction workers is as follows:

based on human body acceleration data acquired by an inertial measurement unit, many researchers have achieved tumble detection algorithms and posture stabilization algorithms. In consideration of the real-time monitoring requirement in the real-time situation, some researchers adopt a Support Vector Machine (SVM) threshold algorithm using a sum acceleration as a detection value, and when the SVM value is larger than a manually set value, the human posture is considered to be suddenly changed, which means a dangerous megaevent or unsafe behavior. The resultant acceleration is calculated as follows:

the SVM threshold algorithm synthesizes the accelerations in all directions into a scalar without directivity, so that the stability of the human posture is represented. SVM threshold algorithms are very effective in detecting static falls (e.g., intentional forward falls in a standing situation), which is common in the elderly or in mobility-impaired patients. However, in a construction environment, similar monitoring algorithms are very effective. In a construction environment, the environment of a worker is more dynamic, the action of the worker is more variable, the worker may get up, sit and rest and bend down many times during work, and if only the worker uses and accelerates, the actions are likely to be recognized as abnormal posture changes. In addition, compared with the normal activities of people in a general environment, the movement state of workers in the operation is also limited by the construction environment, and severe movements such as running and the like are generally avoided. Most related algorithms cannot be directly migrated to be used in a construction environment, and in the current similar research, the overhigh false alarm rate of the algorithms when workers normally work is a great problem at present.

Vectorized acceleration is important in more accurately identifying the type of abrupt attitude change. For example, normal rising and staggering forward may have similar resultant acceleration values, but the former acceleration is mainly reflected on the z-axis perpendicular to the ground, and the latter is reflected on both the x-axis parallel to the ground and the z-axis perpendicular to the ground. Based on this idea, researchers have explored from different directions: for example, an association classification algorithm (FBPAC), mapping the triaxial acceleration data into a bit pattern through an artificially set threshold, and further mining the frequent pattern in the bit pattern to improve the identification accuracy; multiple acceleration sensors are used instead of a single acceleration sensor to enhance the immunity of the system and reduce the false alarm rate. However, the above algorithm still has two problems: (1) the threshold needs to be set in balance between the false alarm rate and the accuracy rate, and the optimal threshold cannot be set in a manual setting mode, so that the generalization of the system is reduced; (2) due to the characteristics of the accelerometer, the return data of the sensor has a great relationship with the spatial position of the sensor relative to the world horizontal coordinate system, and the relative movement of the sensor and the human body in the experimental process can have a great influence on the accuracy of the data. These will be solved in the algorithm proposed by the present design.

The algorithm of the embodiment of the present invention is shown in fig. 3, and the algorithm receives motion sensor data as input, and performs spatial transformation first to maintain spatial invariance of the sensor data. The data stream is then sampled frame by frame in a time window, and 7 features of acceleration and component acceleration variance and head angle pose are extracted for each frame of data. And inputting the features into a pre-trained decision tree for classification, and finally determining whether abnormal actions occur or not by using a continuous event frame judging mechanism so as to judge the occurrence of the dangerous million collision event.

(1) Data registration and transmission

JY901B is in serial port communication with raspberry pi through I2C. The connection mode is shown in table 1, and table 1 is a JY901B and raspberry pi connection mode table.

TABLE 1

After connection, look at the hardware address, default to 0x 50. The acceleration data is transmitted in a 16-ary manner, and each data is transmitted in a lower byte and a higher byte in turn, for example, the X-axis acceleration is Ax, wherein AxL is the lower byte and AxH is the higher byte. The numerical formula for converting the acceleration data into decimal representation in standard units is as follows:

a_x(AxH < 8) | AxL)/32768 × 16g (g is gravity acceleration, and may be 9.8m/s²)

The attitude angle data is also transmitted according to the same rule and scale, and the conversion formula is as follows:

angle_x＝((anglexH＜＜8)|anglexL)/32768*180°

the coordinate system used during all data settlement is the northeast coordinate system, the module is placed in the direction shown in fig. 4, the X axis is arranged rightwards, the Y axis is arranged upwards, and the direction of the spiral thumb of the right hand is the Z axis.

(2) Spatial transformation

The coordinate system used when the sensor settles the data is the northeast coordinate system, which remains relatively stationary with the module. However, the action of the worker can be regarded as occurring in the geodetic horizontal coordinate system, and the measurement of the action of the worker by using the module in the northeast coordinate system will cause an error of the data because the coordinate system of the module cannot be consistent with the coordinate system of the worker, as shown in fig. 5, wherein the triaxial acceleration in the triaxial coordinate x ' y ' z ' is the measurement value desired by the embodiment of the present invention, but the module returns the triaxial acceleration value in the triaxial coordinate xyz. In addition, the accelerometer is always affected by the acceleration of gravity when measuring acceleration. Both of these problems will be solved by the spatial transformation algorithm.

The module attitude angle (i.e., θ in FIG. 5) is known_x,θ_y,θ_z) Thereafter, a spatial transformation matrix may be constructed:

the coordinate system of the module and the acceleration under the geodetic horizontal coordinate system have the following relations:

a_module＝a_x,a_y,a_z]^T

a_world＝a′_x,a′_y,a′_z]^T

a_world＝W·a_module-[0,0,g]^T

on the basis of the acceleration of the head of the worker in the geodetic horizontal coordinate system, the influence of local gravity is subtracted, so that three-axis acceleration values which are not influenced by the gravity can be obtained, and the three-axis acceleration values represent the movement acceleration of the head of the worker in three-axis directions in the geodetic horizontal coordinate system. After the space changes, the obtained acceleration data can eliminate the influence of gravity acceleration and is not influenced by the space position of the sensor, so that the method is more stable and accurate.

(3) Time window sampling

The use of a time sampling window tends to improve the classification of the model when processing streaming data. Because the duration of a destabilizing action event is generally about 0.5 to 1 second, a time window of 0.5 seconds is good for sampling, and 5 frames of data can be sampled in the time window.

(4) Feature extraction

The SVM value is still a valid feature for discriminating a sudden change in human posture and is thus also calculated as one of the features of the input model. In order to make full use of the data within the time window, the variance of the acceleration data within the time window is calculated as input data to the decision tree. Furthermore, head angle data (i.e., attitude angle) is also introduced as an important feature. Through transverse comparison, the characteristic combination of the acceleration variance and the angle original data is selected to perform best in the algorithm.

All feature calculations are shown in fig. 6. Sensor initial data including only three-axis accelerationFrom the three-axis angle, the angle data is not processed, i.e. theta is reserved_x,θ_y,θ_zThree features. For triaxial acceleration, a resultant acceleration calculation formula is used for calculating the resultant acceleration, namely SVM, frame by frame, and then the variance value of the partial acceleration and the resultant acceleration from the current frame to the previous five frames is calculated in a time window of 0.5s, so that the four acceleration variance value characteristics are obtained. And finally, inputting the seven characteristics into a discriminant model.

(5) Decision tree model discrimination

In order to solve the defect that the threshold needs to be manually set in the traditional algorithm, a machine learning method needs to be introduced. Theoretically, machine learning methods suitable for classification can be applied to similar monitoring systems, including k-nearest neighbors, decision trees, naive bayes, logistic regression, support vector machines, artificial neural networks, and the like. Considering that the data used by the algorithm has the characteristics of uneven distribution, continuity and few dimensions, and the processing capacity of raspberry group Zero is limited, the decision tree is the most suitable machine learning model for the system.

The decision tree model is represented as a tree structure. Each node in the model means that the model judges the dataset about a certain attribute, forming two branches (in the case of a regression tree, multiple branches may be formed), wherein each branch outputs the result judged according to the node attribute. And (4) sequentially iterating, and finally, when no branch exists, each leaf node of the decision tree represents a classification result, as shown in fig. 7.

Decision trees belong to supervised learning, and a model needs to be trained on a data set with known classification results to classify the data set with unknown classification results. Enough data are collected through one-time pre-experiment, and classification results are judged manually, so that a decision tree model is trained.

The principle of dividing data by a decision tree is to change the data from unordered to ordered, and the information entropy is used for describing the ordered degree of the data. After each branching, the information entropy of the data set should be reduced as a basis for dividing the data. When each attribute is judged, the change amount of the information entropy after the data set is divided according to the attribute is calculated, and the change is described by using a relevant calculation mode. After each division, the information entropy of the data set needs to be guaranteed to be reduced to the maximum extent. Different algorithms use different bases to divide features, ID3 uses information gain obtained by calculating information entropy, C4.5 uses information gain ratio to divide features based on ID3, and CART uses a Gini coefficient to divide. Practice proves that CART has better performance on solving the task of classifying large samples, and meanwhile, the branch form of the CART is a simple binary tree form and conforms to the requirement of system classification, so the CART is adopted in the system.

The decision tree model receives characteristic combination input and outputs binary judgment: whether the current signature is consistent with the signature of the critical megainstability event. The binary 0 and 1 represent mismatch and coincidence, respectively.

(6) Continuous event frame discrimination

Decision tree models cannot be accurate to one hundred percent over all inputs. Inevitably, a small fraction of normal actions will be identified as actions due to a dangerous collision and vice versa. For the case where the dangerous mega-collision event frames are identified as normal event frames, it should be noted that a dangerous mega-collision event duration often exceeds 0.5 seconds, which also means that at least 5 event frames are included, and the probability of actions that they are all identified as normal is negligible. For the former, even if the model can achieve 99% of accuracy, the time occupied by normal actions is far longer than the action time caused by critical collision, a large number of false alarms still occur, so that field managers are difficult to judge the real critical collision event, which is the biggest disadvantage of similar research in application at present. And recognizing and inspecting decision tree judgment results of all event frames in the current time window by the continuous event frames, and if all the decision tree judgment results are true, determining that the dangerous megabyte collision event really occurs, thereby greatly reducing the false alarm rate. As shown in fig. 8, the discrimination result of each frame in a segment of normal walking data stream is shown. Wherein the small circle is a time window, however, it contains only one event frame identified as a risk mega-collision event frame, and therefore, the risk mega-collision event is not considered to actually occur; the large circle is also a time window that contains all event frames that are identified as a jeopardy crash event frame, and therefore the actual occurrence of a jeopardy crash event is considered.

The sensor-based automatic identification device 10 for the megaevent of instability of a construction worker is further described below by way of specific examples, as follows:

in order to quantitatively evaluate the effect of the algorithm and the device, a simulation experiment is designed to test the algorithm and the device. The experiment is carried out in an indoor simulation scene, the plane operation condition of a common construction site is simulated, and testers need to wear the system provided by the invention to carry out the experiment. Meanwhile, a camera is additionally arranged to shoot the whole experiment process so as to record the time point of each dangerous omen collision event.

Three instability conditions are mainly simulated in the experiment and can be summarized into three conditions:

(1) instability due to active factors: workers are hit directly by impacts, thrown materials, tools, horizontally moving hoists, misplaced ladders or dangerously;

(2) instability due to passive factors: instability caused by the fact that a worker does not pay attention to road conditions or obstacles, the worker collides or is dangerous to collide with a fixed object in a construction scene, the worker is stumbled or is dangerous to stumble by the obstacles on the ground, and the worker steps off;

(3) instability due to overhead impact: workers are concentrated or are at risk of being hit by falling material or tools above.

For the three main destabilizing categories, the experiment will simulate the destabilizing event in different ways. It should be noted that the following simulation method does not necessarily completely recover the instability event actually occurring in the construction scene, but the main purpose is to reproduce the movement state and posture change of the personnel when the risk megainstability event occurs.

(1) Instability due to active factors: the researcher can manually charge and impact the moving tested person from the back and the left and right horizontal directions, and the tested person should keep balance as much as possible. The situation simulates the condition that the constructor is impacted by instruments, building materials, tools, hoisted objects and other personnel during operation;

(2) instability due to passive factors: the investigator will suddenly block the traveling subject with soft material from the front, and the subject should avoid hitting the obstacle as much as possible. Such scenarios simulate situations where the constructor hits or is at risk to hit a fixture on the construction site due to various conditions during operation;

(3) instability due to overhead impact: the investigator will throw a brick-sized soft object over the person under test. Such scenarios simulate situations where a constructor is hit by an object falling from the air or is at risk during work.

In an experiment, a human subject is required to follow a fixed path. The travel path includes four turns to simulate normal turn-around actions of constructors during travel. During the advancing process, the event can occur in different modes on the premise that the tested person is not informed of the occurrence of the event, so as to simulate different dangerous megainstability events in a real construction scene. Each experiment comprises 3 rounds, and each round can simulate the 3 types of the dangerous megainstability events in an interpenetration mode, so that after the same action is repeated for multiple times, a tested person can presume the type of the next simulation event. Meanwhile, in the experiment, the tested person is also required to make routine actions such as stopping, bending, squatting and the like according to requirements during the traveling process, and the actions are not recognized as a dangerous omen destabilization event.

Bubble has been laid at the experimental site and has been bathed the cushion, simultaneously in the experimentation, need be tried on and put on a whole set of motion protective equipment, the protection tester does not receive the collision injury. Researchers need to pay close attention to the reaction being tested during the course of the experiment. The soft object for blocking is a foam plate, and the soft object thrown upwards is a sandbag with the weight of about 500g, so that the safety helmet is harmless to a human body wearing the safety helmet. The overall experimental scheme is shown in fig. 9, and the flow is as follows:

the tested person is required to keep normal walking in the experimental process, and if any change occurs in the normal walking state, the normal walking state is recovered from the end of the normal walking state, and the interval time of the change is considered to contain a complete action.

In the experiment, experimenters are required to record the whole process. The data acquired by the sensor can be directly stored in the local storage of the raspberry pie in real time, and the data can be downloaded to a computer for analysis through the FTP (File Transfer Protocol).

Two people were invited to participate in the experiment and they were informed of the steps and possible scenarios of the experiment. After the necessary protective device is worn, the experiment starts, and the experimental flow is basically consistent with the preliminary experiment.

The data collected by the verification experiment is in the form of the current system identification result and the time stamp at a certain moment. The researchers need to record the actions of the tested persons and the current condition of the system indicator lights in turn during the experiment. An exemplary experimental record is shown in table 2. Wherein, table 2 is an experimental record table.

TABLE 2

Where o represents the indicator light turning on as expected: for normal actions of squatting, bending down and sitting down, the indicator light is turned on in green; the indicator light is required to be red for abnormal actions of collision, blocking and falling objects. If an error is indicated ● is recorded. After the data recording is completed, the researcher also needs to compare the live video to ensure that no data is missed or misread. The verification experiment finally collects 119 action data and corresponding system indication conditions, including 45 normal actions, including 23 squat actions and 22 bending actions. There were 74 abnormal events, including 30 active impact events (impacts), 18 passive impact events (obstructions), and 26 overhead impact events (crashes), distributed similar to the frequency distribution of the threat impact events in the example study. In the normal advancing process, false alarms do not occur in starting, turning and stopping, and the false alarms are not recorded according to the experimental rules. Specific data are shown in table 3, wherein table 3 is a table of verification experiment results.

TABLE 3

Further, the effect analysis is as follows:

(I) analysis of accuracy, recall and precision

The evaluation criteria for characterizing the classifier mainly include Accuracy (Accuracy), Precision (Precision) and Recall (Recall). The accuracy rate describes the proportion of all samples in which the prediction result is the same as the sample, the precision rate describes the frequency value predicted to be correct in the example of the positive sample, and the recall rate describes the frequency predicted to be correct in the example of the positive sample. Generally, the higher the precision ratio, the more accurate the algorithm is; the higher the recall, the more sensitive the algorithm. The calculation method of the accuracy, precision and recall is shown in the following formula:

in the verification experiment record data, the 'correct feedback' simultaneously comprises two situations of TP and TN, wherein TP corresponds to the frequency of responding to the dangerous mega-instability event, TN corresponds to the frequency of not responding to the normal event, FP corresponds to the frequency of identifying the normal event as the dangerous mega-instability event, and FN corresponds to the frequency of not responding to the dangerous mega-instability event. The accuracy of the algorithm for all events can thus be calculated, as shown in table 4. Wherein, table 4 is an accuracy table of event identification.

TABLE 4

Recall and precision are meaningless for normal events, because for normal events, the algorithm should not look out these events as abnormal events, and TP does not contain the response result of the system to normal events. Therefore, the precision and recall of the algorithm for all the hazard collision events are calculated separately, and the result is 97% and 87.8%, respectively.

(II) false alarm rate analysis

The research additionally pays attention to the false alarm rate of the method in identifying the dangerous million destabilizing events, because the biggest obstacle of hindering the research result from being applied is the high false alarm rate in the current similar research. The excessively high false alarm rate causes the algorithm to frequently identify normal events as abnormal events, and rather, the algorithm brings great trouble to security management.

The false alarm rate is a rate of the events that are actually normal among all the events discriminated as abnormal events. The calculation method is shown as the following formula:

in the experiment, the recorded normal actions are mainly of two types: squat and bend over. Although the actions to be tested for voluntary, such as normal walking, small-step fast walking, turning, stopping, starting, and the like, belong to normal actions, the actions are not recorded because the actions are not designed in advance and no alarm is triggered in the experimental process. And calculating the relevant indexes of the verification experiment data according to a false alarm rate calculation formula to obtain the overall false alarm rate of the system, which is 3.0%.

(III) analysis of real-time Properties

In the hardware design, an LED lamp is arranged on the safety helmet as an indicator lamp. By observing the change of the LED indicator light, the response result of the system to the current event can be directly seen. By manually comparing the videos, the interval between the system response and the event occurrence can be directly observed and obtained. The video is imported into adobe remiere and viewed frame by frame. When the indicator light changes, a label is marked on a time axis; similarly, when an abnormal event occurs (as a sign of occurrence when an external force acts on a human body), a label is also placed on the time axis. And calculating the average interval between the two types of labels to obtain the average response time of the system, and taking the average response time as the basis of real-time analysis. The missed exception event, as well as the normal action that triggered the alarm, are not recorded. Table 5 gives an example of a portion of the time stamp data therein. Wherein, table 5 is an example table of time tag records of event occurrence and system response.

TABLE 5

Statistics show that all response times are within 1 second. By calculation, the average response time of the system was 0.43 seconds.

The verification proves that the effectiveness of the algorithm monitoring the critical megainstability event in the embodiment of the invention is proved, and the accuracy, precision ratio and recall ratio of the algorithm should be concerned at the same time for the judgment of an application algorithm.

The overall accuracy of the algorithm is 90.3%, which is lower than the overall accuracy of the decision tree model directly used in the pre-experiment on the verification data set, because the observation target in the verification experiment is not a separate data frame but an event, and normal actions such as walking and the like are not recorded. On the specific abnormal action caused by the danger million collision event, the algorithm can achieve the precision ratio of 97.0 percent and the recall ratio of 87.8 percent, wherein the recall ratio is basically consistent with the algorithm before the judgment of the unadditized continuous event frame, and the precision ratio is greatly improved.

The recall ratio of the algorithm is 87.8%, which shows that the algorithm has a sensitive response to the occurrence of the detection of the critical instability event. The recall ratio is basically consistent with that obtained by directly using the decision tree model, which shows that the decision tree model is a main bottleneck for improving the sensitivity of the model to the risk events. The accuracy of the decision tree model depends to a large extent on the size of the training data and the accuracy of the labeling. On the premise of ensuring the accuracy of the labeling, due to the influence of epidemic situations, enough suitable personnel cannot be found for the experiment, and the lack of pre-training data obviously causes certain adverse influence on the performance of the decision tree model.

The false alarm rate of the algorithm is only 3%, which is superior to some approximate researches, which shows that the algorithm has better tolerance on normal rest and posture adjustment actions possibly occurring in the operation of constructors, and fully shows the effectiveness of a continuous event frame discrimination mechanism in improving the tolerance of the algorithm on the normal actions.

The average response event of the algorithm is 0.43 seconds, and the maximum response time does not exceed 1 second. The algorithm can monitor the occurrence of the accident in a very short time after the dangerous million destabilizing accident occurs, and sends out a warning through the indicator, which means that the research proposes the computing capability of hardware composition, and can be competent for the task of monitoring the dangerous million collision accident in real time when driving the real-time recognition algorithm.

In five scenes designed in experiments, the algorithm is better in the four scenes of squat, bending down, collision and falling, and is more general in the blocking scene. It is observed that the handling performance of the tested object is different because the blocking does not directly act on the tested object, wherein part of the handling actions may not be executed in the preliminary experiment. This also shows the importance of an effective, comprehensive training data set on the machine learning class model.

According to the sensor-based automatic construction worker instability risk megaevent recognition device provided by the embodiment of the invention, based on the monitoring of the stability of the posture of the human body, the construction environment and the operation mode are considered, so that the occurrence of instability can be effectively monitored in real time under the condition of not influencing the operation of workers, and the device has high expansibility.

The method for automatically identifying the instability risk megaevent of a construction worker based on a sensor, which is provided by the embodiment of the invention, is described next with reference to the attached drawings.

FIG. 10 is a flow chart of a method for automatically identifying a destabilizing megaevent for a construction worker based on sensors in accordance with one embodiment of the invention.

As shown in fig. 10, the method for automatically identifying the instability megaevent of the construction worker based on the sensor comprises the following steps:

in step S101, event identification data is collected;

in step S102, when it is identified that a destabilization risk megaevent occurs according to the event identification data, a destabilization risk megaevent alarm is issued.

It should be noted that the foregoing explanation of the embodiment of the sensor-based device for automatically identifying a megaevent of instability of a construction worker also applies to the method for automatically identifying a megaevent of instability of a construction worker based on a sensor in this embodiment, and details are not repeated here.

According to the method for automatically identifying the instability risk megaevent of the construction worker based on the sensor, which is provided by the embodiment of the invention, the instability action can be effectively monitored in real time under the condition of not influencing the operation of the worker by taking the construction environment and the operation mode into consideration on the basis of monitoring the stability of the posture of the human body, and the method has high expansibility.

In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or N embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.

Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "N" means at least two, e.g., two, three, etc., unless specifically limited otherwise.

Claims

1. A construction worker instability risk megaevent automatic identification device based on a sensor is characterized by comprising:

a carrier body for carrying the identification device;

the acquisition device is used for acquiring event identification data;

an external prompter; and

and the controller is used for controlling the external prompter to send out a destabilization risk megaevent alarm when the occurrence of the destabilization risk megaevent is recognized according to the event recognition data.

2. The apparatus of claim 1, wherein the event identification data comprises one or more of three-axis acceleration, three-axis angles, three-axis magnetic data, and air pressure values.

3. The apparatus of claim 1, further comprising:

and the power supply equipment is used for supplying power to the controller.

4. The device of claim 1, wherein the external cue comprises at least one acoustic alarm device and at least one optical alarm device.

5. The apparatus of claim 1, further comprising:

and the communicator is used for sending the event information of the instability risk megaevent to a preset terminal.

6. The apparatus of claim 1, further comprising:

and the recording device is used for recording the field data of the constructors.

7. The apparatus according to claim 1 wherein the carrier body is on a wearable device, wherein the wearable device comprises a hard hat, a belt, and glasses.

8. A construction worker instability risk megaevent automatic identification method based on a sensor comprises the following steps:

collecting event identification data;

and when the instability risk megaevent is identified according to the event identification data, sending out an instability risk megaevent alarm.

9. The method of claim 8, wherein the event identification data comprises one or more of three-axis acceleration, three-axis angle, three-axis magnetic data, and air pressure values.

10. The method of claim 8, further comprising:

and sending the event information of the instability risk megaevent to a preset terminal.