KR20110121885A - Apparatus and method for traffic accident recognition - Google Patents

Abstract

PURPOSE: A traffic accident recognizing device and a method thereof are provided to apply a rule-based determination technique to image information and sound information, thereby reducing the amount of calculation. CONSTITUTION: A road information collecting unit(10) collects image information and sound information of a road. An accident information extracting unit(20) extracts the accident object image information except an image about a background from the image information. The accident information extracting unit extracts the sound information of a crash sound which is generated by a traffic accident from the sound information. An object information pre-processor(30) generates motion pattern information. An accident determining unit determines whether the motion pattern information is about a traffic accident by applying a preset rule to the motion pattern information.

Description

Apparatus and Method for Traffic Accident Recognition

The present invention relates to a traffic accident recognition technology at an intersection, and more particularly, to analyze and process image information collected from a camera and acoustic information collected from a microphone to calculate a motion vector of an object (person or vehicle), The present invention relates to a traffic accident recognition device for automatically recognizing a traffic accident at an intersection from an exercise state and a method of recognizing a traffic accident using the same.

Generally, CCTVs for monitoring traffic conditions and accidents and cameras for speed monitoring are installed throughout the road.However, monitoring these cameras in real time requires effort and time. It was difficult to respond in real time, and recorded image information was often used to identify the cause after an accident occurred.

Therefore, recently, research or patents have been submitted for automatically recognizing traffic accidents by analyzing images, analyzing traffic sounds at the time of an accident, and analyzing the amount of vibration or impact when a traffic accident occurs.

Briefly introducing the related patent literature,

The Korea Electronics and Telecommunications Research Institute combines the internal condition of the vehicle (temperature, speed, brake, airbag operation, impact) + external traffic information (real-time traffic information-traffic jams, traffic accident information, GPS location information, dangerous time zone information) We have proposed an invention for a system that alerts the current vehicle to the risk of accidents. (A traffic accident alarm device and method, application number: 2007-0123048)

Deechi Co., Ltd. proposed a method of detecting an accident by using a vibration sensor on a road, by using a vibration amount when a vehicle passes normally and a vibration amount when an accident occurs. (Ubiquitous traffic accident detection system, application number: 2007-0107240). However, this patent document does not provide a standard for the normal amount of vibration and the amount of vibration in the event of an accident.

Jung Yang-kwon, a personal inventor, sensed the speed through GPS to determine if he was overspeed and warned, and suggested a service to dispatch the police after judging that there was a traffic accident when it was too low compared to the surrounding traffic conditions or speed limit. Continuous Flow Accident Prediction System Using Mobile Communication Terminal, Application No. 2007-0056613) In this patent document, only the speed of GPS is used as the basis of traffic accident discrimination.

Based on the assumption that the sound produced in a collision is different from the everyday sound, Kim Jae-I, a personal inventor, proposed a method of identifying an accident by analyzing the type, size, and frequency distribution of the sound. Application number: 2007-0060475)

Hyun Un-hyuk, a personal inventor, has proposed a system that recognizes and alerts to crimes and traffic accidents using image analysis (especially image analysis through differential image) (automatic security system for crime and accident prevention using computer image analysis technology, Application number: 2008-0040227)

In addition, many individual inventors and companies are working on ways to recognize traffic accidents in real time. However, most of them use simple rule-based algorithms or already developed image analysis / acoustic analysis techniques, except for rule-based algorithms for recognizing traffic accidents based on information obtained through image analysis or acoustic analysis. It is not presented separately.

In addition, in spite of these studies, there is a problem that does not suggest a method on how to fuse the information when a plurality of cameras or microphones obtains a large amount of information on a single accident site.

In view of solving the above problems, the present invention obtains information on the direction and speed of movement of a vehicle and a person (hereinafter referred to as a customer body) from image information and sound information collected through a video camera installed at an intersection. The first technical problem is to provide a traffic accident recognition apparatus and method by applying a hierarchical decision method to information.

In addition, the present invention provides a traffic accident recognition apparatus and method that can install a plurality of video cameras at various points of the intersection in addition to the first technical problem to fuse the various information collected at the same time to increase the accuracy of traffic accident recognition. Making provision is the second technical problem.

However, the technical problem of the present invention is not limited to the above-mentioned matters, and other objects not mentioned will be clearly understood by those skilled in the art from the following description.

In order to achieve the above technical problem, a traffic accident recognition apparatus according to the present invention includes a road information collection unit for collecting image information and sound information of a road, and customer image information other than a background image of the image information. An accident information extracting unit for extracting sound source information of a collision sound generated by a traffic accident among the sound information, and object information for generating motion pattern information which is information about a change amount of a motion vector of a customer in the customer image; It includes a pre-processing unit and an accident determination unit to determine whether or not the traffic accident situation by applying a predetermined rule to the exercise pattern information, and determine the traffic accident situation by performing a probability judgment based on a probability model if it is determined that the traffic accident situation. do.

Here, the accident information extracting unit extracts the image information of the accident object, which is the region of interest, by applying a predetermined algorithm to the image information of the road and extracts the image information of the accident object at the accident spot using the collision sound. It is preferable to include a sound source information extracting unit for extracting the sound source information that is information about the distance between the generated sound source and the traffic accident recognition device and the direction in which the sound source is generated.

The sound source information extracted by the accident information extracting unit may be obtained in consideration of the degree of inconsistency with a predetermined or larger sound or a predetermined frequency pattern among the sound information.

In addition, the accident determination unit,

The sound source information extracted from the accident information extraction unit and the object when the rule-based determination unit and the rule board determination unit determine a traffic accident situation by applying a predetermined rule to the exercise pattern information. It is also preferable to include a probability-based determination unit for determining whether the traffic accident situation by applying a probability model to the exercise pattern information generated by the information preprocessor.

Also preferably, the road information collecting unit may include a video data collecting unit collecting and recording image information of the road to an external server, and an acoustic data collecting unit collecting and recording sound information of the road to the server.

In addition, the rule-based determination unit may generate a result of calculating the change amount of time of the exercise pattern information generated by the object information preprocessor to determine that the traffic accident situation when the result exceeds a predetermined reference value.

The probability-based determination unit may use a Bayesian network as the probability model.

In addition, in order to achieve the above technical problem, a plurality of traffic accident recognition apparatus is provided to be installed at a predetermined position according to the present invention, the road information collection unit for collecting the image information and sound information of the road, the image information of the An accident information extractor configured to extract video information of the customer body except for the background image, and extract sound source information of a collision sound generated by a traffic accident among the sound information; In the object information preprocessor for generating the exercise pattern information which is information on the amount of change, the information fusion unit for generating the fused information by applying a predetermined algorithm to the exercise pattern information generated in the object information preprocessor and the information fusion unit Determining whether or not a traffic accident is applied by applying predetermined rules to the generated converged information, If the judge determines that the situation underwent a probability based on a probability model should include a determination of the accident determined whether the accident situation.

Here, the accident information extracting unit applies a predetermined algorithm to the image information and extracts the image information of the customer body, which is a region of interest, and a sound source generated at the accident spot using the collision sound. And a sound source information extracting unit for extracting sound source information which is information on a distance between the traffic accident recognition apparatus and a direction in which the sound source is generated.

Here, the information fusion unit forms the motion vector information generated by the object information preprocessor of the plurality of traffic accident recognition apparatuses as motion vectors of the absolute coordinate system, and generates a fused motion vector by combining the motion vectors of the absolute coordinate system. Good to do.

The motion vector of the absolute coordinate system is formed by using sound source information extracted from the accident information extracting unit, the vertical angle of the road information collecting unit, the front direction of the road information collecting unit, and the horizontal angle of the sound source. The fused motion vector is preferably generated by combining the motion vector of the absolute coordinate system with weights of constants that are inversely proportional to the distance between the sound source included in the sound source information and the road information collection unit.

On the other hand, in order to achieve the above technical problem, the traffic accident recognition method according to the present invention, (a) collecting the image information and sound information of the road, and (b) except for the background image of the image information Extracting an image of the customer body image, and extracting sound source information of a collision sound generated by a traffic accident among the sound information; and (c) an exercise pattern information which is information about a change amount of the exercise vector of the customer body in the customer body image information. (D) determining a traffic accident situation by applying a predetermined rule to the exercise pattern information generated in step (c), and (e) determining that the traffic accident situation is in step (d). And determining the traffic accident situation by applying a probabilistic model to the sound source information extracted in the step (b) and the exercise pattern information generated in the step (c).

Here, in step (d), it is preferable to generate a result of calculating the change amount of time of the exercise pattern information to determine the traffic accident situation when the result exceeds a predetermined reference value.

In addition, the probabilistic model of step (e) may use a Bayesian network.

Meanwhile, a traffic accident recognition method using a plurality of traffic accident recognition apparatus installed at a predetermined position includes (a) collecting image information and sound information of a road; and (b) an image related to a background of the image information. Extracting the customer information except for the customer, and extracting sound source information of the collision sound generated by the traffic accident from the sound information; and (c) an exercise which is information on the amount of change of the motion vector of the customer in the customer information; Generating pattern information; (d) generating a fused information by applying a predetermined algorithm to the exercise pattern information, and (e) applying a predetermined rule to the fused information generated in the step (d). Judging the traffic accident situation by applying and (f) if it is determined that the traffic accident situation in step (e), the sound source information extracted in step (b) and the luck generated in step (d) Determining whether the traffic accident is by applying a probability model to the pattern information.

Here, the step (e) comprises the step (e1) of forming the motion pattern information as the motion vector of the absolute coordinate system and (e2) generating the fused motion vector by combining the motion vectors of the absolute coordinate system formed in the step (e1). It may also include a step.

In addition, it is preferable that the probabilistic model of the step (f) uses a Bayesian network.

According to the present invention grasped from the contents of the present specification, information on the direction and speed of movement of a vehicle and a person (hereinafter referred to as a customer body) is obtained from image information and sound information collected through a video camera installed at an intersection. By applying hierarchical decision-making techniques to information, traffic accidents can be recognized, thus reducing the time and effort required to monitor traffic accidents. To prevent traffic congestion.

In addition, according to the present invention, if the rule-based determination technique is not applied to the acquired image information and sound information, the probability-based determination technique is not applied when a predetermined condition is not satisfied, and thus a large amount of computation and memory are not required.

In addition, since the present invention forms information collected by multiple video cameras into a single piece of fused information and applies a hierarchical decision method to it, it is possible to recognize a traffic accident more accurately than using a single information.

1 is a block diagram schematically showing an apparatus for recognizing a traffic accident according to an embodiment of the present invention;
2 is a view for explaining the operation of the traffic accident recognition apparatus according to an embodiment of the present invention,
3 is a diagram illustrating a motion vector of a customer in a region of interest;
4 is a diagram illustrating a process of calculating a motion vector difference between objects in the object information preprocessor of the present invention;
5 is a view illustrating a process of applying a hierarchical determination technique to determine a traffic accident situation according to an embodiment of the present invention;
6 is a view illustrating an example of using a Bayesian network as a probability-based determination method in a traffic accident recognition apparatus according to an embodiment of the present invention.
7 is a block diagram schematically showing a traffic accident recognition apparatus according to another embodiment of the present invention;
8 is a view illustrating a state in which information is obtained using a traffic accident recognition apparatus according to another embodiment of the present invention;
FIG. 9 is a view illustrating a process of forming various pieces of information collected by a traffic accident recognition apparatus according to another embodiment of the present invention into one fused information;
10 is a flowchart illustrating a method of recognizing a traffic accident according to the present invention.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the description herein, when a component is described as being connected to another component, this means that the component may be directly connected to another component or an intervening third component may be interposed therebetween. First, in adding reference numerals to the components of each drawing, it should be noted that the same reference numerals are assigned to the same components as much as possible, even if shown on different drawings. At this time, the configuration and operation of the present invention shown in the drawings and described by it will be described as at least one embodiment, by which the technical spirit of the present invention and its core configuration and operation is not limited.

As used herein, the term 'unit' refers to a hardware component such as software or a field programmable gate array (FPGA) or an application specific integrated circuit (ASIC), and performs certain roles. But wealth is not limited to software or hardware. The 'unit' may be configured to be in an addressable storage medium and may be configured to execute one or more processors. As an example, a 'part' may include components such as software components, object-oriented software components, class components, and task components, and processes, functions, properties, procedures, subroutines. , Segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. In addition, the functionality provided in each component may be combined into a smaller number of components or further separated into additional components.

Hereinafter, specific details will be described through embodiments of the present invention. Advantages and features of the present invention and methods for achieving them will be apparent with reference to the embodiments described below in detail with the accompanying drawings.

1 is a block diagram schematically showing a traffic accident recognition apparatus according to an embodiment of the present invention.

As shown in FIG. 1, the traffic accident recognition apparatus 100 according to an embodiment includes a road information collector 10, an accident information extractor 20, an object information preprocessor 30, and an accident determiner 40. ).

The road information collection unit 10 is provided in the traffic accident recognition apparatus 100 installed at any point of a place where a vehicle and a person, such as a road or an intersection, can move to collect information to be collected and are provided with image information and sound information ( S _in ). Specifically, the road information collecting unit 10 may be implemented as any device capable of monitoring the road situation, such as a video camera, a CCTV, a speed monitoring camera, a computer, a digital camera, and a sensor node. Therefore, the following may be referred to as a camera or a video camera in place of the term road information collection unit, it should be noted that this should not be construed as a camera only. In addition, the information collected by the road information collection unit 10 is recorded and stored as data in a server connected to the outside by wireless or wired, and transmits the information so that it can be searched if a future search is required.

The accident information extracting unit 20 is provided to extract information related to a desired traffic accident from various image information and sound information collected by the road information collecting unit 10. The accident information extracting unit 20 may collect information collected by the road information collecting unit 10 in real time in real time, and may include image information of a general traffic flow, road noise, vehicle horn sound, etc., rather than a traffic accident situation. It is intended to use only properly filtered information. That is, the accident information extracting unit 20 classifies and extracts the customer body image information from the acquired image information and the background and foreground image information. Here, an algorithm for distinguishing a background and a foreground of an image by using a difference between frames of an image may be used, and such an algorithm may be a known and predetermined algorithm. For example, an object tracking method using GMM can be used.

The accident information extracting unit 20 removes noise from the acquired acoustic information. That is, the collision sound accompanying the accident is extracted from the acoustic information. Here, a method of analyzing based on the frequency, pattern, and size of the sound may be used. Through this method, the distance between the sound source of the accident spot and the traffic accident recognition apparatus 100 and the direction in which the accident spot is located based on the traffic accident recognition apparatus 100 can be known.

The object information preprocessor 30 generates the exercise pattern information by analyzing the exercise pattern of the four-customer body extracted by the accident information extractor 20. In the present invention, the exercise pattern information is generated by using the movement direction and the speed of the four-customer body. Create In other words, by analyzing the motion vector of the customer body, the movement direction and the size of the customer body, the amount of change according to the time of the motion vector is calculated and provided to the accident determination unit 40 to be described later.

The accident determination unit 40 applies the hierarchical determination technique to the motion pattern information of the customer customer provided by the object information preprocessor 30 and the sound source information extracted from the accident information extraction unit 20. It determines whether it is a traffic accident situation and generates it as an output signal (S _out ). Here, the hierarchical judgment technique means that the hierarchical judgment is carried out when the predetermined criteria are satisfied in the judgment of the lower hierarchy. Applying such a hierarchical determination technique, if the requirements of the lower hierarchy are not satisfied, it is not judged whether the requirements of the upper hierarchy are satisfied, and there is an advantage of not using a large amount of computation and memory. On the other hand, if the requirements of the lower class are satisfied, it is determined whether the upper class meets the requirements again, so that the traffic accident can be recognized with higher reliability. That is, in case of rule judgment, since no learning is required, a large amount of data is not required for actual application, while in case of probability-based judgment, it can be applied even when a part of the data is lost. Through the use of this hierarchical structure, the disadvantages of both can be compensated for.

Detailed configuration and operation of the accident determination unit 40 and the detailed operation of each unit according to an embodiment of the present invention will be described later with reference to FIG.

2 is a view for explaining the operation of the traffic accident recognition apparatus according to an embodiment of the present invention.

As shown in FIG. 2, the traffic accident recognition apparatus 100 according to an embodiment includes a video data collector 110, an acoustic data collector 120, and a region of interest (ROI) extractor 130. ), The sound source information extractor 140, the object information preprocessor 150, the rule-based determiner 160, and the probability-based determiner 170.

The video data collector 110 and the acoustic data collector 120 are used to record video and audio information related to an accident. First, as described above, the video data collecting unit 110 may be configured as a video camera or a CCTV as a part of collecting image information S _{in. 1} for recognizing an accident. Image information collected here is recorded in a database of an external server in real time and stored for later retrieval.

The sound data collector 120 collects sound information (S _{in .2} ) related to an accident and may be configured as a microphone attached to a video camera or a microphone for collecting sound. Typically, the sound data collector 120 and the video data collector 110 may be integrally formed as one device or may be formed in the form of a microphone built in a camera.

The ROI extractor 130 is used to distinguish the customer (vehicle, person) from the collected image information. An algorithm for distinguishing the background and the foreground of the image by using the difference between the frames of the image (eg object tracking using GMM) Etc.) may be used. In addition, the region of interest extractor 130 provides basic information of objects to be used in the object information preprocessor, and the algorithm used may be a published algorithm or program. (E.g. OpenCV, etc.)

The sound source information extracting unit 140 is for distinguishing a collision sound associated with a traffic accident among the sound information collected by the sound data collecting unit 120, and a method of analyzing the sound source information extracting unit 140 may be used based on the frequency, pattern, and size of the sound. To analyze, extract the distance and direction of the sound source. That is, the voice information extracting unit 140 determines the sound volume related to the accident with respect to a predetermined loudness or acoustic frequency pattern through a simple rule, and roughly estimates the distance and direction of the sound source. Algorithms related to this may be used as disclosed algorithms or programs, and thus description thereof will be omitted.

The object information preprocessor 150 uses the ROI information extracted from the image information to analyze the directions and speeds of the clients and return the results to the rule-based determination unit 160 and the probability-based determination unit 170. To pass. That is, the object information preprocessor 150 calculates the movement direction and size of the object and the amount of change of the movement for the rule-based determination unit 160 and the probability-based determination unit 170 for determining the traffic accident. The items processed by the object information preprocessor 150 are summarized in the following [Table 1].

Preprocessing items details Conflict check of ROI Check whether there are overlapping ranges of ROI areas of different objects. If there is an overlapping area, it means that two objects are located in the short distance, which increases the probability of an accident. Calculation of motion vector by object Calculate the degree to which the X and Y coordinates of the center point of the given object's ROI area change from frame to frame. This gives you an overview of the speed and direction of the object. Of motion vector
Calculate Difference When an accident occurs, the objects related to the accident change the direction and speed of movement greatly. Therefore, after calculating the difference of the motion vectors between two objects involved in the accident, the change amount (increase or decrease) is reflected in the accident inference.

3 and 4 exemplarily illustrate a process of generating motion pattern information of an object to be executed by the object information preprocessor 150. That is, FIG. 3 is a diagram illustrating a motion vector of a customer in a region of interest, and FIG. 4 is a diagram illustrating a process of calculating a motion vector difference (V1-V2) between objects in the object information preprocessor of the present invention. to be.

3 and 4, the object information preprocessor 150 generates the motion pattern information using the difference of the motion vectors from the motion vector change of the object 1 and the object 2 to the rule-based determination unit 160 to be described later. to provide.

The rule-based determining unit 160 determines whether the current traffic accident has occurred based on a predetermined rule, and if it is determined that the traffic accident is a traffic accident, the rule-based determining unit 160 makes a decision based on the probability model to increase the accuracy. Leave the final judgment. As described above, according to the present invention, the decision method of the hierarchical structure, that is, the rule-based decision unit 160 which is the lower layer, is used to determine the probability-based determination unit 170 which is the upper layer.

In other words, although probability-based methods are better suited to deal with thinking in uncertain real-world situations, probability-based algorithms require a lot of computation and memory, making it difficult to use probability-based algorithms every frame, whereas rule-based methods guarantee accuracy. Because you can't. Therefore, in this case, the two-level hierarchical structure is used to first determine whether the current traffic accident situation is performed through the rule-based method, and when it is determined to be similar to the traffic accident situation, a more accurate judgment result is obtained by using a probability-based method.

Hereinafter, a process of applying the hierarchical determination method will be described in detail with reference to FIG. 5.

5 is a diagram illustrating a process of applying a hierarchical determination technique to determine a traffic accident situation according to an embodiment of the present invention.

Referring to FIG. 5, the rule-based determination unit 160 records and uses a change amount of a motion vector difference between objects as a feature value. In general, when a traffic accident occurs, the objects (vehicles and people) related to the traffic accidents change their direction and speed greatly. Therefore, if there is a significant difference between the exercise state of the previous state and the exercise state of the current state, it can be assumed that the accident occurred. Based on the difference in the motion vectors between the objects provided by the object information pre-processing unit 30 calculates the amount of change over time, if the value exceeds the predetermined reference value (a) it is determined as a traffic accident.

Next, the probability-based determination unit 170 considers the collision check result of the region of interest (whether the region of interest overlaps), the difference in the amount of change in the motion vector, the speed of the object, the direction of the object, whether the movement of the object continues, etc. Construct a probabilistic model, and return the probability that the accident occurred based on the collected information as a value between 0 and 1. There are various methods in the probabilistic model such as Hidden Markov Model, Conditional Random Field, Bayesian network, but it is preferable to use Bayesian network here. Bayesian network is a method of modeling the relationship between variables in the form of a graph. In the process of probabilistic inference, the number of variables increases and the computational amount increases exponentially. give. FIG. 6 is a diagram illustrating such an example, and illustrates an example of using a Bayesian network as a probability-based determination method in a traffic accident recognition apparatus according to an exemplary embodiment of the present invention.

Until now, a method of recognizing a traffic accident based on information collected from a single camera and a microphone has been described. However, in the case of an intersection, several cameras and microphones may be installed depending on an angle. In this environment, the information collected from the cameras can be integrated to achieve higher accuracy than using only the information collected from each camera. Here, a traffic accident recognition apparatus according to another embodiment of the present invention for fusing information in such a case will be described with reference to FIGS. 7, 8, and 9.

FIG. 7 is a block diagram schematically illustrating an apparatus for recognizing a traffic accident according to another exemplary embodiment of the present invention. FIG. 8 is a view for explaining how information is acquired using the apparatus for recognizing a traffic accident according to another exemplary embodiment of the present invention. FIG. 9 is a view illustrating a process of forming various pieces of information collected by a traffic accident recognition apparatus according to another embodiment of the present invention into one fused information.

As shown in FIG. 7, the traffic accident recognition apparatus 700 according to another embodiment includes N road information collecting units 700, an accident information extracting unit 20, an object information preprocessor 30, and an information fusion unit. 70 and the accident determination unit 40. Here, the descriptions of the accident information extracting unit 20, the object information preprocessor 30, and the accident determining unit 40 overlap with those described above, and thus description thereof will be omitted.

N road information collecting units 10 are provided with a plurality of road information collecting units as described in an embodiment to collect image information and sound information (Sin) from places where vehicles and people pass, such as intersections or roads. do. The plurality of road information collectors are arranged and installed at a predetermined location, and in a practical embodiment, they may be installed in structures such as traffic lights, roadside trees, and streetlights at intersections.

The information fusion unit 70 generates fused information by applying a predetermined algorithm to the exercise pattern information generated by the object information preprocessor 30, which will be described in detail below.

8 shows the situation at the intersection with several cameras and microphones. When multiple cameras are installed as described above, when an accident is photographed by multiple cameras, an algorithm capable of combining the information from the multiple cameras may be needed. In this case, it is assumed that the acoustic information exists and the distance and angle from the acoustic information to the scene of the accident can be determined.

In order to synthesize the image information, the data from the camera are combined based on the distance and angle of the accident site and the camera is converted into absolute coordinate data and then combined. The process consists of two steps:

(Step 1) Change the direction of the motion vector using the angle of the scene where the accident occurred in the camera

Given the angle of the camera and the scene of the accident, it is primarily transformed into a coordinate system taken from the front of the camera. Second order transformation is performed by considering the angle of the camera itself. Finally, the motion vector of the world coordinate system is calculated. The calculation proceeds as shown in Equation 1 below.

[Equation 1]

은 카메라1(11)에서 관측된 객체의 운동벡터,

은 카메라1(11)과 사고지점과의 수직방향 각도,

은 카메라1(11)의 정면방향과 사고지점과의 수평방향 각도,

는 절대좌표값으로 변환된 운동벡터를 의미한다. 여기서는 예시적으로 카메라1(11)의 경우만을 설명하였으나, N개로 구비된 카메라 모두에 대하여 같은 방법으로 계산이 진행됨은 자명하다.
here,

Is the motion vector of the object observed from camera 1 (11),

Is the vertical angle between the camera 1 (11) and the accident location,

Is the horizontal angle between the front direction of the camera 1 (11) and the accident location,

Denotes a motion vector converted to an absolute coordinate value. Here, only the case of the camera 1 (11) is described as an example, but it is obvious that the calculation is performed in the same manner for all the cameras provided with N.

(Step 2) Size conversion of motion vector using distance from camera to accident spot

In order to fuse the magnitudes of the motion vectors captured by each camera, they are combined by weighting constants that are inversely proportional to the distance from each camera to the point of accident. This progresses as shown in Equation 2 below.

[Equation 2]

,

,

는 카메라1,2,3에서 사고지점까지의 거리에 반비례하는 상수로

+

+

=1.0임,

는 카메라 1(11)에서 관측된 운동벡터의 각도변환 운동벡터,

는 카메라 2(12)에서 관측된 운동벡터의 각도변환 운동벡터,

는 카메라 3(13)에서 관측된 운동벡터의 각도변환 운동벡터,

은 융합된 운동벡터이다. here,

,

,

Is a constant that is inversely proportional to the distance from the camera 1, 2, 3 to the accident

+

+

= 1.0,

Is the angular transformation motion vector of the motion vector observed from Camera 1 (11),

Is the angular transformation motion vector of the motion vector observed from Camera 2 (12),

Is the angular transformation motion vector of the motion vector observed by camera 3 (13),

Is the fused motion vector.

Here, an example of using three cameras has been described, but the number of terms of Equation 2 may be variously changed according to the number of cameras.

In this way, when motion vectors of objects obtained from N cameras are combined, finally, the combined information is used to recognize a traffic accident situation. 9 is a view schematically showing the overall process for this. That is, a combination of weighted values of coordinate system transformed motion vectors is formed to form fused information, and the traffic accident situation is recognized using the information. The process of determining whether or not a traffic accident using the converged information is applied to the above-described hierarchical determination technique, and thus no duplicate description will be given here.

Next, a description will be given of a method for recognizing a traffic accident according to an embodiment of the present invention and another embodiment.

10 is a flowchart illustrating a method of recognizing a traffic accident according to the present invention.

As shown in FIG. 10, the traffic accident recognition method includes collecting image information and sound information (S10), extracting image information of a customer's body, and extracting sound source information (S20), and an exercise pattern of the customer's body. Analyzing (S30), applying the rule-based determination to the exercise pattern information and the sound source information (S40), and applying the probability-based determination using the probability model (S50). The description of the method of recognizing a traffic accident made by these steps will be clearly understood with reference to the couple who described the traffic accident recognition apparatus according to an embodiment of the present invention, and thus the description thereof is omitted for the sake of brevity. .

As described above, the present invention has been described by way of limited embodiments and drawings, but the present invention is not limited to the above-described embodiments, which can be variously modified and modified by those skilled in the art to which the present invention pertains. Modifications are possible. Accordingly, it is intended that the scope of the invention be defined solely by the claims appended hereto, and that all equivalents or equivalent variations thereof fall within the spirit and scope of the invention.

10: road information collection unit 20: accident information extraction unit
30: object information preprocessing unit 40: accident judgment unit
110: video data collection unit 120: sound data collection unit
130: region of interest extraction unit 140: region of interest extraction unit
160: rule-based decision unit 170: probability-based decision unit

Claims (17)

In the traffic accident recognition device,
A road information collector configured to collect image and sound information of a road;
An accident information extracting unit extracting customer information except for a background image from the image information, and extracting sound source information of a collision sound generated by a traffic accident among the sound information;
An object information pre-processing unit for generating exercise pattern information which is information about a change amount of the motion vector of the customer body in the customer body image information; And
A traffic accident including an accident determination unit that determines whether or not the traffic accident situation by applying a predetermined rule to the exercise pattern information, and determines whether or not the traffic accident situation by performing a probability judgment based on a probability model when the traffic accident situation is determined. Recognition device. According to claim 1, wherein the accident information extraction unit
A region of interest (ROI) extraction unit for extracting image information of an accident object which is a region of interest by applying a predetermined algorithm to the image information of the road; And
And a sound source information extraction unit for extracting sound source information which is information on a distance between the sound source generated at the accident site and the traffic accident recognition device and the direction in which the sound source is generated using the collision sound. The method of claim 1,
The sound source information extracted by the accident information extracting unit is obtained in consideration of the degree of inconsistency with a predetermined size or more than a predetermined size of the sound information or a predetermined frequency pattern. The method of claim 1, wherein the accident determination unit,
A rule-based decision unit which determines a traffic accident situation by applying a predetermined rule to the exercise pattern information; And
If it is determined that the traffic accident situation is determined by the ruler determining unit, the probability information is determined by applying a probabilistic model to the sound source information extracted by the accident information extraction unit and the exercise pattern information generated by the object information preprocessor. Traffic accident recognition apparatus comprising a probability-based determination unit. According to claim 1, wherein the road information collecting unit
A video data collection unit for collecting the image information of the road and recording it to an external server; And
And a sound data collecting unit for collecting the sound information of the road and recording the same to the server. The apparatus of claim 4, wherein the rule-based determination unit
And generating a result of calculating a time-based change amount of the exercise pattern information generated by the object information preprocessing unit and determining that the traffic accident situation occurs when the result exceeds a predetermined reference value. The method of claim 4, wherein the probability-based determination unit
Traffic accident recognition apparatus using a Bayesian network as the probability model. In the traffic accident recognition device provided with a plurality to be installed in a predetermined position,
A road information collector configured to collect image and sound information of a road;
An accident information extracting unit configured to extract image data of the customer body except for the background image from the image information, and extract sound source information of a collision sound generated by a traffic accident among the sound information;
An object information pre-processing unit for generating exercise pattern information which is information about a change amount of the motion vector of the customer body in the customer body image information;
An information fusion unit generating fused information by applying a predetermined algorithm to the exercise pattern information generated by the object information preprocessor; And
Determine whether the traffic accident situation by applying a predetermined rule to the converged information generated by the information fusion unit, and if it is determined that the traffic accident situation is determined to determine whether the traffic accident situation by performing a probability judgment based on a probability model Traffic accident recognition device comprising a determination unit. The method of claim 8, wherein the accident information extraction unit
A region of interest (ROI) extracting unit for extracting image information of a customer, which is a region of interest, by applying a predetermined algorithm to the image information; And
And a sound source information extraction unit for extracting sound source information which is information on a distance between the sound source generated at the accident site and the traffic accident recognition device and the direction in which the sound source is generated using the collision sound. The method of claim 8, wherein the information fusion unit
The motion vector information generated by the object information preprocessor of the plurality of traffic accident recognition apparatuses is formed as a motion vector of the absolute coordinate system, and the motion vector of the absolute coordinate system is combined to generate a fused motion vector. Incident Recognition Device. The method of claim 10,
The motion vector of the absolute coordinate system is formed by using the sound source information extracted from the accident information extracting unit, the vertical angle of the road information collecting unit, the front direction of the road information collecting unit, and the horizontal angle of the sound source,
The fused motion vector is generated by combining a motion vector of the absolute coordinate system with a weight value of a constant inversely proportional to the distance between the sound source included in the sound source information and the road information collecting unit. Device. In the recognition method of traffic accidents,
(a) collecting image information and sound information of a road;
(b) extracting customer image information except for a background image from the image information, and extracting sound source information of a collision sound generated by a traffic accident from the sound information;
(c) generating motion pattern information which is information about a change amount of the motion vector of the customer customer in the customer customer image information;
(d) determining a traffic accident situation by applying a predetermined rule to the exercise pattern information generated in step (c); And
(e) If it is determined that the traffic accident situation in step (d), whether the traffic accident situation by applying a probabilistic model to the sound source information extracted in step (b) and the exercise pattern information generated in step (c) Recognizing a traffic accident comprising the step of determining. The method of claim 12, wherein step (d)
And generating a result of calculating a change amount of time of the exercise pattern information to determine a traffic accident situation when the result exceeds a predetermined reference value. The method of claim 12,
The probability model of step (e) is a traffic accident recognition method characterized in that using the Bayesian Network (Bayesian Network). In the traffic accident recognition method using a plurality of traffic accident recognition apparatus installed in a predetermined position,
(a) collecting image information and sound information of a road;
(b) extracting customer image information except for a background image from the image information, and extracting sound source information of a collision sound generated by a traffic accident from the sound information;
(c) generating motion pattern information which is information about a change amount of the motion vector of the customer customer in the customer customer image information;
(d) generating a fused information by applying a predetermined algorithm to the motion pattern information;
(e) determining a traffic accident situation by applying a predetermined rule to the fused information generated in step (d); And
(f) If it is determined that the traffic accident situation in step (e), whether the traffic accident situation by applying a probabilistic model to the sound source information extracted in step (b) and the exercise pattern information generated in step (d) Recognizing a traffic accident comprising the step of determining. The method of claim 15, wherein step (e)
(e1) forming the motion pattern information as a motion vector of an absolute coordinate system; And
(e2) generating a fused motion vector by combining the motion vectors of the absolute coordinate system formed in step (e1). 16. The method of claim 15,
The probability model of step (f) is a traffic accident recognition method characterized in that using the Bayesian Network (Bayesian Network).

Images