KR101095528B1 - An outomatic sensing system for traffic accident and method thereof - Google Patents

Abstract

A surveillance camera which photographs and monitors a surveillance region of a continuous flow road; An image acquisition unit for converting a real-time video photographing the surveillance region with a surveillance camera into digital image data; An image traffic information collecting unit for analyzing the digital image received from the image obtaining unit to calculate an amount of traffic information and collecting the calculated traffic information data; Analyzing digital images received from the image acquisition unit automatically detects the traffic congestion rate, traffic accidents and accidents in the surveillance area, alerts you when an accident or accident is detected, or releases an accident or accident. An image information processing analysis unit which performs a function of automatically releasing an alarm situation when the image is processed; Collect real-time data to collect the traffic information data processed by the video traffic information collection unit and send it to the traffic control center, and send the detection result of accidents and accidents detected by the video information processing and analysis unit to the traffic control center. Or a system control unit for receiving and processing a data retrieval request or a remote control command of a camera from a traffic control center; A data storage unit configured to store a real-time digital image input through the system controller and to configure and operate data acquired through the system controller as a database; And transmits traffic information data from the system control unit to the traffic control center, accident and accident detection result alarms and alarm release data, and stored video data, or receives data request commands or control commands from the control center to the system control unit. Disclosed is a continuous flow road accident and sudden automatic detection detection system comprising a unit.

Continuous flow road, surveillance area, traffic information, accident and accident, accident analysis filter, exception handling, accident judgment

Description

Continuous outflow road accident detection system and its method {An outomatic sensing system for traffic accident and method

The present invention relates to an automatic discrimination detection system and a method for detecting an accident of a vehicle driving on a continuous flow road or occurrence of an accident on the road, and more particularly, to a failure, an accident and an accident of a vehicle occurring on a road using an image processing algorithm. A system and method for automatically determining and detecting a situation occurrence.

As the industry develops, the number of vehicles, which are the means of transportation, increases, the traffic volume increases rapidly, and the speed of the roads progresses. Accidents occurring on the road are not limited to the problem between the perpetrator and the victim, and the aftermath of the accident also affects subsequent vehicles. Such aftereffects are inevitably larger as the number of vehicles driving on the road and the speed of vehicle flow are faster. If the duration of an accident in any one lane on a three-lane road exceeds 10 minutes, the aftermath may extend up to 10 km back. Therefore, the early detection of accidents or accidents on the road can be said to have a great influence on traffic communication because it enables the rapid handling of accidents and accidents. The automatic detection system for accidents or accidents occurring on these roads can not only significantly reduce traffic and congestion time, but also induce decentralization of vehicles concentrating on the accident area through detours. There is an advantage.

As a method of detecting accidents and accidents on the roads on the road, algorithms such as APID, McMaster, DES, and DELOS are used as traffic engineering methods. It collects traffic information such as traffic volume and occupancy time in real time, calculates moving average of this traffic information for every 20 ~ 80 second time period, and applies it as the main variable of the algorithm. However, these algorithms collect traffic information such as loop detectors, magnetic detectors, laser detectors, or image detectors between 500m to 2km long sections to collect real-time traffic information. It is a method of detecting the change of traffic flow by comparing with the amount of information in a large amount of traffic information database stored in advance and tracking the changes of macro traffic volume to detect the aftermath of an accident or accident. It is a way to detect the occurrence of accident in the section. However, there are long detection times to track changes in traffic information and to estimate accidents or outbreaks, and problems that are indistinguishable from congestion in long congestion sections other than actual traffic accidents, and traffic accidents in situations where there is not much traffic. The problems are very difficult to figure out without the driver's report.

On the other hand, as a detection technology for accidents occurring at intersections, an automatic accident system at the intersection is used to generate events through the sound of accidents occurring at the time of the accident, and to record and store certain time before and after the accident in the video storage device. Has been developed and used in Japan and the United States, but it is not only difficult to detect accidents when the sound is not collected, but also automatically records accidents in relatively limited areas such as intersections, and analyzes the results of the accidents. There is a positive effect that provides the function to clearly identify the victims, but it automatically detects the obstacles of traffic flow due to single accident or breakdown of the vehicle that does not generate an accident sound, so that accidents and accidents necessary for the rapid removal of the obstacles are eliminated. Not enough to provide a means of early detection of the situation Information is. In particular, as a means of early detection of accidents on roads with a longer continuous section than on intersections such as national highways and highways, it is very difficult to collect the sound of an accident at the time of an accident. There is a problem that can not use the means to detect the occurrence of an accident using.

In addition to the above-mentioned macro-accident and accident detection algorithms for accidents occurring on continuous flow roads, real-time monitoring of restricted areas is performed in real-time and automatically detects accidents and accidents in case of traffic flow obstacles. There is a need for means to monitor incidents and incidents so that they can be effectively and quickly eliminated, minimizing additional human and economic losses in the aftermath of the incident.

The present invention not only automatically detects an accident in a continuous flow road using image processing technology, but also detects an accident regardless of the presence or absence of an error sound at the time of an accident, as well as in a much larger area than an intersection. Continuous flow road accident and accident detection system and detection method that can monitor the change status of traffic flow in real time, detect traffic information in real time, and effectively detect traffic congestion and provide congestion status information. The purpose is to provide.

Continuous flow road accident and sudden automatic detection detection system of the present invention for achieving the above object comprises a surveillance camera for monitoring by monitoring the monitoring area of the continuous flow road; An image acquisition unit for converting the real-time video captured by the surveillance camera into digital image data; An image traffic information collecting unit for analyzing the digital image transmitted from the image obtaining unit to calculate the amount of traffic information and collecting the calculated traffic information data; Analyzes the digital image received from the image acquisition unit to automatically detect the traffic congestion rate, traffic accident and accident situation in the surveillance area, issue an alarm when an accident and accident situation are detected, or release the accident and accident situation. An image information processing and analyzing unit for automatically releasing an alarm condition when the alarm is released; Collects the traffic information data processed by the video traffic information collecting unit and collects real-time data to transmit to the traffic control center, and detects the result of the accident and accident detected by the video information processing and analyzing unit at the traffic control center. A system control unit configured to send a data request or receive a data retrieval request from a traffic control center or a remote control command of a camera; A data storage unit for storing a real-time digital image input through the system control unit and configuring and operating data acquired through the system control unit as a database; And transmits traffic information data from the system control unit to the traffic control center, accident and accident detection result alarms and alarm release data, and stored video data, or receives a data request command or control command from the control center to the system control unit. And a data transmission unit.

Here, the video traffic information collecting unit analyzes the received digital image to measure the traffic volume, vehicle speed and occupancy time of each vehicle on the continuous flow road, and the speed moving average of each lane for a certain period and during the period. It is desirable to calculate the sum of the traffic volume of the vehicle and the occupancy time of the vehicle passing through the surveillance region during the period.

In addition, the image information processing analysis unit, a congestion rate processing unit for calculating the spatial occupancy and congestion rate of the vehicles collected by the image traffic information collection unit; On the basis of the information collected by the video traffic information collecting unit, it detects the accident signs area from the driving state of the vehicles in operation, and detects the occurrence of accidents and accidents on the road by continuously monitoring the detected accident signs area. An accident and accident detection process for performing a first-step accident detection; An accident analysis filter unit configured to determine whether or not a precise accident has occurred in the vehicle and the area detected by the accident and accident detection processor unit at the first stage; An exception processing unit that performs exception processing on misrecognition factors such as stagnation state, diffuse reflection of light, and environmental factors that are inaccurate factors among the judgment results of the accident analysis filter unit; And an accident final decision control unit which finally determines an accident determination for an accident situation finally confirmed through exception processing in the exception processing unit, and automatically releases the accident situation when the accident situation is released.

The congestion rate processor extracts an image of vehicles in the surveillance area, analyzes the space occupancy rate and the amount of space movement of the vehicles, calculates the congestion rate, and continuously monitors the change in the congestion rate. When the above congestion occurs, it is better to generate an accident preliminary alarm due to congestion.

The accident and accident detection process unit generates a background image from every input image frame, extracts a difference image between the background image and the input image, extracts an image of a predetermined threshold or more as an image of the vehicle, and extracts the image of the extracted vehicle. It is good to analyze the change of the vehicle image every image frame, and to analyze the vehicle and the stop state of the accident for a predetermined time to perform the first preliminary detection of the accident.

In addition, the accident analysis filter unit analyzes the first accident preliminary accident and sudden situation occurrence area for a predetermined time from the vehicle and the incident image of the preliminary accident situation detected by the accident and accident detection process unit and the area change rate of the corresponding area and It is good to analyze the change rate and the movement of the surrounding vehicles to filter out the situation, not the accident.

In addition, the accident final decision control unit stores the accident image portion to prevent the process of the stationary accident image is gradually transformed into the background image in the process of repeating the background image generation process through the Gaussian complex model, the input image and the stored accident Using the normalized correlation (NCC) to calculate the correlation of the image part, the existence of the accident vehicle is confirmed, and whether the vehicles on the back of the accident image change lanes and move to the left and right directions of the accident image, and After confirming whether there is a blank area without vehicle image in front of the image, check the inherent existence of the video and the blank area in front of the image to confirm the continuous existence of the accident area. If any of the above conditions are not issued, the alarm is determined to be an accident release situation and the alarm is released. Good to do.

In addition, the continuous flow road accident and sudden automatic detection detection method for achieving the above object, in the method for monitoring the monitoring area of the continuous flow road to automatically detect the vehicle accident and sudden situation, the continuous flow road using a camera Real time shooting; Converting the image photographed by the camera into digital image data; Analyzing the digital image data to calculate a traffic information amount and collecting the calculated traffic information data; Determining whether to execute an accident and accident detection process by comparing the amount of traffic information calculated and collected in the collecting step with a threshold value; When judging as the detection process execution status in the judging step, the digital image received from the image acquisition unit is analyzed to automatically detect the congestion rate, traffic accidents and accidents in the surveillance area, and detect accidents and accidents. A situation detection step of releasing an alarm when an alarm is issued, or releasing an accident or accident; A storage step of storing the collected traffic information data and the accident and accident detection information; And transmitting the collected traffic information data and the accident and accident detection information to a traffic control center. Characterized in that it comprises a.

The situation detecting step may include: calculating a congestion rate according to the quantity and density of vehicles in the surveillance region by analyzing the digital image; Analyzing the digital image to preliminarily determine the types of accidents and accidents on the road; A filtering step of filtering the non-accident situation by analyzing the vehicle and the unexpected image of the pre-determined accident situation detected in the first preliminary determination step for a predetermined time; Analyze the vehicle and accidental images of the pre-determined accident situation detected in the first preliminary judging stage for a certain period of time, and analyze the obstacles that cause the misunderstanding of the accidents such as the influence of the surrounding environment or the diffuse reflection on the road surface at night during rainy weather. An exception handling step of filtering out false detections; Continuously monitoring the remaining images filtered in the filtering step and the exception processing step among the images detected in the first preliminary judging step to determine a final accident and an accident situation; And generating an alarm signal when judging by the final accident and the accident situation.

In the congestion rate calculation step, the congestion rate of the vehicles in the monitoring area is calculated to continuously monitor the change in the congestion rate, and when the calculated congestion rate is determined to be greater than or equal to a preset threshold congestion rate, an accident preliminary alarm is performed. Good to generate.

In addition, the first preliminary judging may include generating a background image from every input image frame; Extracting a difference image between the background image and the input image, and extracting an image having a predetermined threshold value from the extracted difference image as an image of the vehicle; Analyzing the extracted image of the vehicle every image frame, and analyzing the stationary state of the vehicle and the sudden state for a predetermined time to determine the first preliminary accident.

In addition, the filtering step may include detecting the movement amount of the object determined as the accident preliminary detection object in the image frame, and excluding it from the accident determination when the rate of change of the detected object is greater than or equal to the reference value; If it is determined that the area of the object extracted from the image frame is overlapped and changed by the movement of another object, if the area of the changed accident preliminary detection object exceeds the reference ratio by more than the area of the original object, excluding from the accident determination; And when there is movement of the vehicle image around the accident preliminary detection object in the image frame, excluding from the accident determination when a new object passes over the stationary accident preliminary detection object image.

The determining of the final accident and the accident may include: a first checking step of checking whether there is no other vehicle passing through the object determined as the accident preliminary detection object; A second checking step of checking whether vehicles behind the accident preliminary detection object change lanes to move left and right of the accident preliminary detection object; A third checking step of checking whether there is a blank area in front of the accident preliminary detection object and if a blank is maintained for a predetermined time; A fourth checking step of storing a background image generated through the Gaussian complex model and confirming the existence of the accident vehicle using a normalized correlation (NCC) for calculating a correlation between the input image and the accident image; And generating an accident and an alarm when the first to fourth confirmation steps are satisfied, and releasing the accident and an alarm when the at least one of the first to fourth confirmation steps is not satisfied. It is good.

According to the continuous flow road accident and accident detection system and method of the present invention, it is possible to collect the traffic information on the continuous flow road in real time in real time, the accident situation caused by congestion situation, as well as occurred in a very low traffic situation Since there is an advantage that can be detected automatically and more accurately and reliably, even a single accident of a vehicle and a failure caused by a failure of the vehicle, there is an advantage that can effectively detect accidents and accidents occurring in the continuous flow road.

With reference to the accompanying drawings will be described in detail the continuous flow road accident detection system and automatic detection detection system and its detection method according to an embodiment of the present invention.

1 and 2, the continuous flow road accident detection and sudden automatic detection detection system according to an embodiment of the present invention, the surveillance camera 20 for monitoring by monitoring the monitoring area 11 of the continuous flow road 10 And, an image acquisition unit 30, an image traffic information collection unit 40, an image information processing analysis unit 50, a system control unit 60, a data storage unit 70, and a data transmission unit 80. It is provided.

The surveillance camera 20 is installed to photograph the surveillance region 11 of the continuous flow road 10, and transmits the captured real-time video to the image acquisition unit 30.

The image acquisition unit 30 converts the real-time video photographed by the surveillance camera 20 into digital image data, and converts the converted digital image data into the image traffic information collecting unit 40 and the image information processing analyzing unit 50. And the system control unit 60.

과, 주기동안의 차량의 통행량과, 주기동안 감시영역을 통과한 차량의 점유시간의 총 합을 계산한다.The image traffic information collecting unit 40 analyzes the digital image received from the image obtaining unit 30 to calculate the amount of traffic information for each lane and collects the calculated traffic information data. In detail, the image traffic information collecting unit 40 analyzes the received digital image for a predetermined period (t) of the traffic volume Mi (t) of the vehicle on the continuous flow road 10 and the vehicle speed Vi ( t)) and occupancy time (Occi (t)) for each vehicle, and the speed moving average for each lane for a certain period (t)

Calculate the sum of the traffic volume of the vehicle during the cycle and the occupancy time of the vehicle passing through the monitoring area during the cycle.

In this case, the predetermined period is a time interval used for extracting traffic information, and may be set to 20 seconds to 1 minute according to the structure, capacity, and geographical condition of the road, and in particular, most roads should be set to 1 minute. .

More specifically, the video traffic information collecting unit 40 compares each of the thresholds with the amount of traffic information calculated for each period and calculates the congestion rate of the surveillance area or performs an accident detection process according to the comparison result. To provide a signal. That is, the image traffic information collecting unit 40 determines whether the space occupancy time Occi (t) calculated for each period is greater than or equal to the threshold occupancy time Thocc, and the vehicle speed Vi (t) is a critical speed Vcon. Compares the result of the calculation and finds a lane whose traffic information exceeds the threshold set for each lane according to the calculated result, calculates the length of the queue of the lane exceeding the threshold value in the monitoring area, and calculates the queue Comparing the lengths and thresholds yields data to be used for preliminary judgments.

As shown in FIG. 2, the image information processing analysis unit 50 includes a congestion rate process unit 51, an accident and accident detection process unit 52, an accident analysis filter unit 53, an exception processing unit 54, and an accident. The final judgment control unit 55 is provided.

The congestion rate process unit 51 analyzes the digital image received from the image acquisition unit 30 and the traffic information amount collected by the image traffic information collection unit 40 to determine the quantity and space occupancy of the vehicle in the vehicle's surveillance area. Calculate the congestion rate according to.

The congestion rate processor 51 continuously calculates the congestion rate at each predetermined period and detects a change in the congestion rate. Then, when a congestion rate occurs above a predetermined threshold congestion rate (standard congestion rate), an accident preliminary warning due to congestion is issued.

The accident and sudden detection process unit 52 is based on the information collected from the image information collecting unit 40 (traffic information amount) and the digital image transmitted from the image obtaining unit 30, from the driving state of the vehicles in operation. It detects the accident sign area and continuously monitors the detected accident sign area to detect the accident of the vehicle and the occurrence of the unexpected situation on the road, and performs the first stage accident detection.

That is, the accident and accident detection processor 52 generates a background image from every image frame transmitted from the image acquisition unit 30, extracts a difference image between the background image and the input image, and displays an image of a vehicle having a predetermined threshold or higher. The image is extracted, and the change of the vehicle image is analyzed for each image frame from the extracted vehicle image, and the first preliminary detection of the accident is performed by analyzing the stop state of the vehicle and the sudden state for a predetermined time.

That is, as will be described in detail later, the accident and accident detection processor 52 may extract an object stopped for 30 seconds or more from the background screen using a GMM algorithm and an ECD detection method. If the object is found to be stopped for more than 30 seconds from the background screen, it is extracted as the first-stage accident preliminary detection object (vehicle) so as to detect the first-stage accident preliminary detection, and the first accident preliminary detection is performed.

The accident analysis filter unit 53 determines whether there is an accident through a precise analysis of the vehicle and the corresponding area detected by the accident and accident detection process unit 52 in the first step. To this end, the accident analysis filter unit 53 calculates the movement rate of the first-stage accident preliminary detection object, the area change rate of the object, and the movement trajectory of the object (vehicle) near the object, and calculates the calculated values. Compared with the reference values, it should be excluded from the accident judgment or maintained. That is, if there is a change in the amount of movement of an object in the surveillance area, the object (vehicle) is excluded from the accident judgment when the change rate is more than the reference value (10%).

In addition, the accident analysis filter unit 53 may change the first-stage accident when the area of the first-stage accident preliminary detection object (area within the image frame) found in the surveillance area is overlapped and changed by the movement of another object (vehicle). If the area of the preliminary detection object increases compared to the original area, it is excluded from the accident judgment.

In addition, when there is a movement of an object (vehicle) around the first stage preliminary detection object, if a new object (object image) passes over the image of the stationary first stage preliminary detection object, it is excluded from the accident judgment. In this way, the accident analysis filter unit 53 analyzes at least three cases, for example, when the vehicle is paused or moves slowly, or overlaps with a shadow, the movement is fine and is extracted through GMM. Phase preliminary detection can filter out all misunderstandings.

The exception processing unit 54 performs exception processing for misrecognition factors such as a congestion state, diffuse reflection of light, and environmental factors, which are factors of inaccurate judgment, among the judgment results of the accident analysis filter unit 53. Specifically, when the vehicle is stalled for a long time (for example, 10 minutes or more), except in case of accident judgment, except in case of rain or night due to optical diffuse reflection of the road surface, indirect light, etc. Changes caused by shadows, roadside streets and surrounding lighting are also excluded from the accident judgment.

The accident final determination control unit 55 finally determines the accident determination for the first stage preliminary detection object (accident situation) that is finally confirmed through exception processing in the exception processing unit 54, and automatically when the accident situation is released. Release the accident situation. That is, the accident final determination control unit 55 finally issues an accident determination and / or an accident / incident alarm through a situation in which misrecognition occurs in the process of continuously monitoring the first stage preliminary detection object (accident situation) for a predetermined time, If the accident situation is cleared, clear the accident alarm.

Here, the final judgment controller 55 stores the accident image part to prevent the accident image from becoming a background image in the processing of the GMM algorithm after a predetermined time of the accident determination, and calculates a correlation between the input image and the stored accident image part. The normalized cross co-relation (NCC) is used to confirm the existence of the accident vehicle. In addition, the final judging controller 55 checks the traffic volume before and after the accident spot (the presence or absence of other vehicle images passing through the first stage preliminary detection object), the presence of a blank area in front of the accident spot, and the traffic flow trace near the accident spot (1 Preliminary accident detection The vehicle finally determines whether an accident is triggered by changing lanes and moving in the left and right directions, or an alarm is issued or any of the above 4 conditions is not satisfied. If not, the alarm will be cleared as an accident release situation.

The system control unit 60 stores the real-time video input from the image acquisition unit 30 in the data storage unit 70, and collects the traffic information data processed by the image traffic information collection unit 40, the traffic control center ( 90) to collect the real-time data to be transmitted to, and transmits the detection result of the accident and accident detected by the image information processing and analysis unit 50 to the traffic control center 90, or from the traffic control center 90 A data retrieval request or a remote control command from the camera 20 is received and processed.

The data storage unit 70 stores the real-time digital image input through the system control unit 60, or configures and operates the data obtained through the system control unit 60 as a database.

The data transmission unit 80 transmits the traffic information data and the accident and accident detection result alarm and the release data of the alarm from the system control unit 60 to the traffic control center 90, or the system control unit from the traffic control center 90. A data request command or a control command to 60 is received.

Signal transmission between the data transmission unit 80 and the traffic control center 90 may be made through a wired or wireless network.

Hereinafter will be described in detail the detection method using the continuous flow road accident and accident detection system according to an embodiment of the present invention having the above configuration.

Referring to Figures 1 to 5 will be described the continuous flow road accident detection and automatic detection method according to an embodiment of the present invention as follows.

First, an area to monitor traffic, that is, a monitoring area, is set on the continuous flow road 10 (S10).

Then, the surveillance camera 20 is installed to photograph the surveillance region 11 set in the continuous flow road 10 (S11).

The video captured by the surveillance camera 20 is transmitted to the image acquisition unit 30 in real time. The image acquisition unit 30 receives a video and converts the video input in real time into digital image data (S12).

)와, 주기동안(t)의 차량의 통행량과, 주기동안(t) 감시영역을 통과한 차량의 점유시간의 총 합을 산출한다.The data converted into digital image data is transferred to the video traffic information collecting unit 40, and the video traffic information collecting unit 40 analyzes the digital video data to calculate traffic volume for each lane for a predetermined period, and the calculated traffic information data. Collect (S13). Specifically, the video traffic information collecting unit 40 analyzes the digital image, and during a predetermined period (t) the traffic volume Mi (t) of the vehicle in the monitoring area 11 and the vehicle speed Vi ( t)), occupancy time (Occi (t)) for each vehicle, and the average speed for each lane (t) for a certain period (t)

), And the sum of the traffic volume of the vehicle during the period t and the occupancy time of the vehicle passing through the monitoring area during the period t.

Subsequently, the image traffic information collecting unit 40 compares the calculated and collected traffic information amounts with respective threshold values (reference values) to determine whether to perform an accident and accident detection process. Specifically, first, it is checked whether the occupancy time Occi (t) calculated for each vehicle exceeds the threshold value Thocc (S14), and when the vehicle occupancy is exceeded, the vehicle speed Vi (t) Further check whether is less than the critical speed (Vconj) (S15). In the step S15, if the vehicle speed Vi (t) is less than the threshold speed Vconj, it may be determined that the vehicle is pushed. If the vehicle speed Vi (t) is abnormal, the vehicle speed Vi (t) may not be accurate. Check if the free speed (FFS) is exceeded (S16). The free speed (FFS) may be, for example, a speed of 80% or more of the maximum allowable speed or a set congestion speed (60Km / s) or more of the road 10 in the corresponding continuous road.

)를 계산한다(S18).On the other hand, when the vehicle speed Vi (t) is less than the free speed FFF or the vehicle speed Vi (t) is determined to be less than the threshold speed Vconj in the step S16. It is recognized that there is a stagnation speed section for each lane, and the speed change rate is calculated during the period (t) (S17), and then the average speed (t) during the period (t) for each lane (

) Is calculated (S18).

)가 음의 임계값(-Th3) 이하인지 비교하고(S19), 음의 임계값(-Th3) 미만인 경우 정체속도가 심화된 것으로 판단할 수 있으며, 음의 임계값(-Th3) 이상인 경우 적정 음의 임계값(-Th)과 양의 임계값(Th3) 사이에서 평균속도(

)가 유지되는지 확인하고(S20), 상기 음의 임계값(-Th)과 양의 임계값(Th3) 사이에서 유지될 경우에는 정체속도가 일정하게 유지되고 있는 상태로 판단할 수 있다.Calculated average speed

) Is compared to the negative threshold (-Th3) or less (S19), if the negative threshold (-Th3) can be determined that the congestion rate is deepened, if the negative threshold (-Th3) or more appropriate The average speed between the negative threshold (-Th) and the positive threshold (Th3)

) Is maintained (S20), and if it is maintained between the negative threshold value (Th) and the positive threshold value (Th3), it may be determined that the stagnation speed is kept constant.

)가 음의 임계값(-Th)과 양의 임계값(Th3) 사이에서도 유지되지 않는 것으로 확인될 경우, 대략 3 내지 5회의 주기(T)별로 평균속도(

)가 누적되어 증가하는 것이 반복되는지 확인하고(S21), 확인결과 그렇지 않은 경우 정체속도가 유지 내지 심화되고 있는 것으로 판단할 수 있다.In addition, as a result of confirming in the step (S20) the average speed (

) Is determined not to be maintained between the negative threshold (-Th) and the positive threshold (Th3), the average speed (about 3 to 5 cycles T)

If the cumulative increase is repeated (S21), and if it is not confirmed, it can be determined that the stagnation speed is maintained or deepened.

)의 확인 결과 정체속도가 유지되거나 심화되고 있는 것으로 확인되면, 타 차로 교통량 변화의 결과(S22)를 전달받아 차로별로 속도(V1(t), V2(t),...Vn(t))를 비교하여 평가한다(S23).Therefore, the average speed

As a result of the check, it is confirmed that the congestion speed is maintained or deepening, and receives the result (S22) of the traffic volume change by other cars, and the speed (V1 (t), V2 (t), ... Vn (t)) for each lane. Evaluate by comparing (S23).

The lowest level (LOS-F) among the service levels (LOS-A, LOD-B, ... LOS-F) of the vehicle speed set for each lane among the vehicle-specific speeds Vn (t) as a result of the comparison in the step S23. If there are lanes less than) (S24), and if present, the number of lanes (one, two, three, ... n) of LOS-F or less is determined (S25).

Subsequently, the queue length is calculated for each lane (S26). By calculating the stagnant lanes in this way and calculating the speed, queue length, etc. for each lane, it is possible to determine the congestion scale in the lateral and longitudinal directions. do. That is, a congestion rate is calculated by driving the congestion rate process unit 51 with data calculated as described above (such as the number of lanes at which a speed below the congestion rate is generated, the queue length for each lane, etc.) (S27). .

When it is confirmed that the calculated congestion rate exceeds the threshold congestion rate Cr-1 (S28), the video traffic information collecting unit 40 issues a preliminary alert (S29).

Subsequently, the image information processing analysis unit 50 performs the automatic accident detection process with the input digital image and traffic information data. In addition, even if the congestion rate in the monitoring area 11 above the predetermined level in step S16, even if the speed of the vehicle exceeds the free speed (FSS), the automatic accident detection process is performed. That is, if an object suddenly stopped on the road due to an accident or an accident occurred during the movement of the vehicle, the accident automatic detection process is started immediately after the object is stopped.

First, the image information processing analysis unit 50 extracts a background image from the input digital image (S30). In this case, the background image may be extracted from the background image by using the GMM algorithm and the ECD detection method.

The GMM algorithm and ECD detection method used at this stage are as follows.

<GMM Algorithm Execution Process>

GMM is a method of modeling each pixel as a mixture of any number of Gaussian distributions. This method uses the online approximation to update the coefficients of the model. By adjusting the values of the coefficients, a background adapted to various environments can be extracted.

에서 화소 {

}의 히스토리는,time

Pixel at

} History,

(1)

(One)

는 영상의 시퀀스(sequence)이다. 각 화소의 최근 히스토리를 K개의 GMM으로 나타내면 다음과 같다.here,

Is a sequence of images. The recent history of each pixel is represented by K GMMs as follows.

,(2)

,(2)

: 시간

에서

번째 가우시안 가중 계수 (

)here,

: time

in

First Gaussian weighting factor (

)

: 시간

에서

번째 가우시안 평균

: time

in

First Gaussian Average

: 시간

에서

번째 가우시안 분산

: time

in

First Gaussian variance

: 가우시안 확률 분포 함수

Gaussian probability distribution function

에 대하여 각 분포의 표준 편차를 구하고, 표준 편차가 2.5이내인 매칭 분포를 찾은 후, 구한 분포를

값이 큰 순서대로 정렬하여 이 중 몇 개의 분포를 영상의 배경으로 간주하고 다음과 같이 가중계수를 갱신한다. Pixel of new frame

Find the standard deviation of each distribution with respect to, find the matching distribution with a standard deviation of 2.5 or less, and then

The values are sorted in order of increasing order, and some distributions are regarded as the background of the image, and the weighting coefficient is updated as follows.

(3)

(3)

: 가중 변수here,

Weighting variables

: 학습률(learning rate),

Learning rate,

: 정합함수(Matching Function) =

: Matching Function =

: 시정수(Time constant)

: Time constant

번째 프레임에서 갱신되어질 가중변수 및 평균, 분산은 다음과 같이 표현된다.

The weight variables, averages, and variances to be updated in the first frame are expressed as follows.

(4)

(4)

(5

(5

(6)

(6)

As described above, the background image is generated using the GMM algorithm, the generated background image is stored, the background image is stored (S31), and if it is not stored, the background image is extracted from the background image buffer (S32). ). Then, by comparing the stored background image with the current frame image input in real time and obtaining an image of the difference, a difference image corresponding to the object Obj can be obtained from the image frame (S33). The method of generating the difference image and extracting the still object information may be performed through the following method.

<Generation of Tea Image and Extraction of Still Object Information>

The difference image is an image obtained by obtaining a difference between the extracted background image and the image of the current frame, and there is a possibility that a vehicle object exists in the large difference. A process of obtaining a binary image of a background image and a difference image by GMM is as follows.

(7)

(7)

,

(8)

,

(8)

,

(9)

,

(9)

,

,

는 각각 시간

에서의 차영상, 입력동영상 프레임, 배경영상이다.here

,

,

Each time

The difference image, the input video frame, and the background image in.

The binary image obtained in the above manner includes the cast shadow of the vehicle and the vehicle object. An embodiment of the present invention is characterized by using an edge magnitude change ratio detection algorithm (ECD) that can reduce misperception of a vehicle due to reflection of a vehicle signal or light due to night or rain. . Non-change (ECD) detection can be understood as follows.

<ECD detection method>

The variance of the borderline size ratio in the ROI where all the borderline ratios are equal and unchanged is defined as follows and the value is zero.

(10)

10

은 관심영역의 크기,

은 경계선 크기의 비의 분산,

은 경계선 크기 비의 평균이다. 따라서, 입력된 영상 시퀀스의 관심영역에 변화가 없을 경우 다음과 같이 재정리될 수 있다.here,

Is the size of the region of interest,

Is the variance of the ratio of the border size,

Is the average of the borderline size ratios. Therefore, when there is no change in the ROI of the input image sequence, it may be rearranged as follows.

(11)

(11)

로 나타내고, 영상 내 차량의 변화가 발생했다면,Set the brightness of the pixels in the input image

If a change in the vehicle in the image occurs,

,(12)

, (12)

(13)

(13)

When the boundary of the image that satisfies the above condition is found, the boundary of the object becomes.

That is, the difference image is acquired through the ECD detection method as described above, and compared with the threshold value among the obtained difference images (S34 and S37), the image or noise that may be misrecognized is filtered, and the image data falling below the threshold value is filtered. Dispose of it.

The ECD detection method is analyzed for a predetermined time to determine the first preliminary accident and the image (S38). Therefore, the background image BG (t-n) and the resultant images R (n) and R (n-1) can be obtained based on the accident detection cycle (for example, 30 second cycle) (S39).

Through the above process, the accident and accident detection process unit 52 may proceed to the accident detection process to obtain an accident preliminary detection object (image).

Then, the first accident preliminary detection analysis process is performed by the accident analysis filter unit 53 and the exception processing unit 54.

That is, first, the accident analysis filter unit 53 determines whether there is an accident through a precise analysis of the vehicle and the area detected by the accident and accident detection process unit 52 in the first step (S40). By the method.

First, the movement amount of the object determined as the accident preliminary detection object in the image frame is detected, and when the rate of change of the detected object is greater than the reference value is excluded from the accident determination (S41). For example, if the rate of change of the detected object is more than 10%, it can be excluded from the accident judgment.

Second, when it is confirmed that the area of the object extracted from the image frame is overlapped and changed by the movement of another object, the accident judgment is made when the area of the changed accident preliminary detection object exceeds the standard ratio compared to the area of the original object. Excluded (S42).

Third, when there is a movement of the vehicle image around the accident preliminary detection object in the image frame, when a new object passes over the stationary accident preliminary detection object image, it is excluded from the accident determination (S43).

After the filtering process (S40) as described above, the data that may be misrecognized in the preliminary processing unit 54 is excluded. That is, after analyzing each image frame for a predetermined time, if it is determined that the accident preliminary detection object is a long-term stagnant object, it is excluded from the accident judgment (S44).

Next, the false detection situation is treated as an exception by optical diffuse reflection and indirect light by rain or illumination (S45).

In addition, even if the processing is ambiguous logically, for example, due to environmental factors such as roadside streets, surrounding lighting, shadows by objects, etc., the accident determination is treated as an exception (S46).

Through the above process, the first stage preliminary detection object is filtered and exception processed, and the remaining data is finally determined to determine whether an accident or an accident occurs (S50).

Prior to finally determining the occurrence of an alarm and issuing an alarm, the following methods are used to prevent the occurrence of false recognition in the process of continuously monitoring the accident for a certain period of time from the result of the judged image processing. When an accident judgment or accident / incident alarm is issued or an accident situation is released, the accident alarm is automatically released.

That is, in order to prevent the accident image from becoming a background image in the process of GMM algorithm after a certain time, the accident image part (ROI) to automatically release the accident situation after a certain time is released. Extract and store by mask processing (S51), and obtain a normal image correlation coefficient (NCC) between a corresponding ROI region of the input video screen and a stored mask image at predetermined time intervals (20 to 30 seconds) (S52). .

Here, the calculation of the ROI region and the NCC calculation procedure are as follows.

<ROI area calculation>

라 하면 관심영역의 마스크

는,Input real-time video space

The mask of the region of interest

Quot;

(14)

(14)

는,Reference ROI Object Area in Input Real-Time Image

Quot;

(15)

(15)

일 경우, 현재 영상중의 ROI Object는Currently input video

If, the ROI Object in the current video is

(16)

(16)

Becomes The normalized redemption coefficient of the ROI of the reference ROI image and the currently input image is calculated as follows.

<NCC calculation result correlation coefficient>

, 입력된 영상의 ROI영역이

일때 정규화 상관계수 계산은 다음 식 (17)과 같다.The ROI area is mxn pixels and the reference mask image is

, The ROI area of the input image

When the normalized correlation coefficient is calculated as in the following equation (17).

(17)

(17)

는 기준 마스크영상

의 평균레벨,

는 입력 영상의 ROI영역

here,

Reference mask image

Average level of,

Is the ROI area of the input image

The average level of is obtained as follows.

(18)

(18)

(19)

(19)

It is checked whether the NCC calculated by the above method exceeds the threshold value Thncc (S53).

5 shows the binarized mask area M (x, y) and the accident image ROI s (x, y), respectively.

In addition, for the final judgment of the accident, after calculating not only the value of the NCC but also the following three final judgment conditions (Racc) (S55), all of the calculation results of the three conditions (Racc) are the threshold value (Thacc) Check if it exceeds (S56), and only when both of the above confirmation process (S53, S56) is satisfied, the final judgment as an accident and accident situation is issued an alarm (S54).

If it is confirmed that any one of the two check processes (S53, S56) is not satisfied, the accident alarm is released (S57).

Meanwhile, the three determination conditions may be set as follows.

First, check whether there is another vehicle passing through the object that is determined as the accident preliminary detection object (available from the traffic volume information before and after the accident point), and second, vehicles behind the accident preliminary detection object change lanes Check whether the accident preliminary detection object moves in the left and right directions (confirmed by the traffic flow trace information near the accident point), and third, whether there is a blank area in front of the accident preliminary detection object and the space continues for a certain time. You can check it.

Figure 1 is a schematic diagram showing a continuous flow road accident and accident automatic detection system according to an embodiment of the present invention.

FIG. 2 is a flowchart illustrating the image traffic information collecting unit of FIG. 1. FIG.

3A, 3B, and 3C are diagrams for explaining a detection method of calculating a congestion rate of a road in an image information processing analyzer and detecting an accident to finally determine whether an accident occurs.

FIG. 4 is a diagram illustrating an example of extracting a corresponding image using a mask for a ROI (region of interest) for NCC (normalized correlation) calculation of FIG. 3C.

5A illustrates a binarized mask image.

Figure 5b is a diagram showing an accident image ROI.

<Description of Symbols for Main Parts of Drawings>

10. Video camera 20. Digital video acquisition unit

30. Data transmission unit 40. Video traffic information collection unit

50..Image information processing and analysis unit 60 .. System control unit

70. Video data storage 80. Data transmission

Claims (13)

A surveillance camera which photographs and monitors a surveillance region of a continuous flow road; An image acquisition unit for converting the real-time video photographing the surveillance area into the digital image by the surveillance camera; An image traffic information collecting unit for analyzing the digital image transmitted from the image obtaining unit to calculate the amount of traffic information and collecting the calculated traffic information data; Analyzes the digital image received from the image acquisition unit to automatically detect the traffic congestion rate, traffic accident and accident situation in the surveillance area, issue an alarm when an accident and accident situation are detected, or release the accident and accident situation. An image information processing and analyzing unit for automatically releasing an alarm condition when the alarm is released; Collects the traffic information data processed by the video traffic information collecting unit and collects real-time data to transmit to the traffic control center, and detects the result of the accident and accident detected by the video information processing and analyzing unit at the traffic control center. A system control unit configured to send a data request or receive a data retrieval request from a traffic control center or a remote control command of a camera; A data storage unit for storing a real-time digital image input through the system control unit and configuring and operating data acquired through the system control unit as a database; And Transmitting traffic information data from the system control unit to the traffic control center, accident and accident detection result alarms and alarm release data, and stored image data, or receiving a data request command or control command from the control center to the system control unit. It includes; data transmission unit, The image information processing analysis unit, A congestion rate processing unit for calculating spatial occupancy and congestion rate of the vehicles collected by the image traffic information collecting unit; On the basis of the information collected by the video traffic information collecting unit, it detects the accident signs area from the driving state of the vehicles in operation, and detects the occurrence of accidents and accidents on the road by continuously monitoring the detected accident signs area. An accident and accident detection process for performing a first-step accident detection; An accident analysis filter unit configured to determine whether the accident is precise for the vehicle and the area detected by the accident and the accident detection process in the first step; An exception processing unit that performs exception processing on misrecognition factors such as stagnation state, diffuse reflection of light, and environmental factors that are inaccurate factors among the judgment results of the accident analysis filter unit; And And an accident final decision control unit which finally determines an accident determination for an accident situation finally confirmed through exception processing in the exception processing unit, and automatically releases the accident situation when the accident situation is released. The congestion rate process unit, Extracts images of vehicles in the surveillance area, analyzes the occupancy rate and space movement of the vehicles, calculates the congestion rate, and continuously monitors the change in the congestion rate, and reserves an accident according to congestion when congestion occurs above a predetermined predetermined congestion rate. Trigger an alarm, The accident and accident detection process unit, A background image is generated from every input image frame by using Gaussian complex model (GMM) algorithm and non-change (EDC) detection method, and the image above the predetermined threshold is extracted as the vehicle image by extracting the difference image between the background image and the input image. And analyzing the change of the vehicle image from the extracted vehicle image every image frame, and analyzing the stationary state of the vehicle and the sudden state for a predetermined period of time, and detecting the first primary accident. Automatic discrimination detection system. The method of claim 1, wherein the image traffic information collecting unit, Analyze the received digital image and measure the traffic volume, vehicle speed and occupancy time of each lane on the continuous flow road, and the average speed moving average of each lane, the traffic volume of the vehicle during the cycle, A continuous flow road accident detection and detection system, characterized in that the sum of the occupancy time of the vehicle passing through the monitoring area. delete delete delete The method of claim 2, wherein the accident analysis filter unit, Analyze the first occurrence of accidents and accidents that were preliminarily determined for a predetermined time from the vehicle and the incident image detected by the accident and accident detection process unit for a predetermined time, and the area change rate and movement change rate of the corresponding area and the movement situation of the surrounding vehicles. Continuous flow detection and accident detection system for continuous flow, characterized in that to filter out the situation, not the accident. According to claim 2, The accident final judgment control unit, a) In the process of repeating the background image generation process through the Gaussian complex model, the accident image part is saved to prevent the background image from being gradually converted into the background image, and the correlation between the input image and the stored accident image part is calculated. Use the normalized correlation coefficient (NCC) to determine the existence of the accident vehicle, b) check whether the vehicles behind the accident image change lanes and move left and right in the accident image, c) Check whether there is a blank area without vehicle image in front of the accident image to check the unique existence of the image and the blank area in front of it. d) A procedure for confirming the continuing existence of the accident area is used to finally determine the accident and issue an alarm, or if any of the conditions a), b), c) and d) does not apply, Continuous automatic road accident detection and accident detection system, characterized in that to release the alarm by judging. In the method of automatically detecting the vehicle accident and accident by monitoring the monitoring area of the continuous flow road, Photographing the continuous flow road using a camera in real time; Converting the image captured by the camera into a digital image; Analyzing the digital image to calculate a traffic information amount and collecting the calculated traffic information data; Determining whether to execute an accident and accident detection process by comparing the amount of traffic information calculated and collected in the collecting step with a threshold value; When judging as the detection process execution status in the determining step, the digital image is analyzed to detect the traffic congestion rate, traffic accidents and accidents in the surveillance area automatically, and issue an alarm when an accident or accident is detected. A situation detection step of releasing an alarm situation upon the release of an accident and accident situation; A storage step of storing the collected traffic information data and the accident and accident detection information; And And transmitting the collected traffic information data and the accident and accident detection information to a traffic control center. The situation detection step, Calculating a congestion rate according to the quantity and density of vehicles in the surveillance region by analyzing the digital image; Analyzing the digital image to preliminarily determine the types of accidents and accidents on the road; A filtering step of filtering the non-accident situation by analyzing the vehicle and the unexpected image of the pre-determined accident situation detected in the first preliminary determination step for a predetermined time; Analyze the vehicle and accidental images of the pre-determined accident situation detected in the first preliminary judging stage for a certain period of time, and analyze the obstacles that cause the misunderstanding of the accidents such as the influence of the surrounding environment or the diffuse reflection on the road surface at night during rainy weather. An exception handling step of filtering out false detections; Continuously monitoring the remaining images filtered in the filtering step and the exception processing step among the images detected in the first preliminary judging step to determine a final accident and an accident situation; And And generating an alarm signal when judging by the final accident and accident situation. In the congestion rate calculation step, the congestion rate of the vehicles in the monitoring area is calculated to continuously monitor the change in the congestion rate, and when the calculated congestion rate is determined to be greater than or equal to a preset threshold congestion rate, an accident preliminary alarm is generated. , The first preliminary determination step, Generating a background image from every input image frame; Extracting a difference image between the background image and the input image, and extracting an image having a predetermined threshold value from the extracted difference image as an image of the vehicle; Analyzing the extracted image of the vehicle every image frame, and determining the first preliminary accident by analyzing the stop state of the vehicle and the sudden state for a predetermined time; continuous flow road accident and sudden automatic Discrimination detection method. delete delete delete The method of claim 8, wherein the filtering step, Detecting an amount of movement of an object determined as an accident preliminary detection object in an image frame and excluding it from an accident determination when a rate of change of the detected object is greater than a reference value; If it is determined that the area of the object extracted from the image frame is overlapped and changed by the movement of another object, if the area of the changed accident preliminary detection object exceeds the reference ratio by more than the area of the original object, excluding from the accident determination; And Using a Gaussian complex model (GMM) algorithm and unchanged (EDC) detection, when a vehicle image moves around an accident preliminary detection object in every input image frame, a new object passes over the stationary accident preliminary detection object image. The step of excluding from the accident judgment if the continuous road accident and accident detection method comprising a. The method of claim 8 or 12, wherein the step of determining the final accident and the accident situation, A first checking step of checking whether there is no other vehicle passing by the object determined as the accident preliminary detection object; A second checking step of checking whether vehicles behind the accident preliminary detection object change lanes to move left and right of the accident preliminary detection object; A third checking step of checking whether there is a blank area in front of the accident preliminary detection object and if a blank is maintained for a predetermined time; A fourth checking step of storing a background image generated through the Gaussian complex model and confirming the existence of the accident vehicle using a normalized correlation (NCC) for calculating a correlation between the input image and the accident image; And If the first to fourth confirmation step satisfies all, generating an accident and accident alarm, and if not satisfying at least one of the first to the fourth confirmation step; releasing the accident and accident alarm; Continuous flow road accident and sudden automatic detection detection method characterized in that.

Images