Radar-based traffic accident identification method and electronic equipment

Abstract

The invention discloses a radar-based traffic accident identification method and electronic equipment. The method comprises the following steps: acquiring traffic data and detection parameters from a radar arranged at a target intersection; wherein the traffic data comprises the reference position and speed of each object detected by the radar, and the detection parameters comprise the boundary value of the detection area of the radar; generating N trajectory diagrams based on the traffic data and the detection parameters; identifying the M track diagrams, and determining whether a traffic accident exists at the target intersection; wherein the M track diagrams are part or all of the N track diagrams; m and N are integers; and if the traffic accident exists at the target intersection, determining the position of an accident point of the traffic accident and a bypass area caused by the traffic accident. And the identification accuracy and speed of the traffic road accident are improved.

Description

Technical Field

[0001] The present application relates to the field of smart city traffic technology, and in particular to a radar-based traffic accident identification method and electronic equipment. Background Art

[0002] In the current urban traffic system, traffic road detection is usually carried out by capturing real-time images and applying image processing and machine learning techniques to identify and handle various obstacles on the road. Among them, image feature method and behavior feature method are two main video analysis technologies. Image feature method can effectively identify the characteristics of accident vehicles, construction vehicles, safety fences and other obstacles; behavior feature method can analyze dynamic behavior patterns in the video, such as vehicle behavior and pedestrian dynamics.

[0003] However, in video analysis technology, general surveillance cameras can only cover a limited area, and it is difficult to effectively monitor accidents or obstacles that are beyond the camera's field of view. In addition, whether it is the image feature method or the behavior feature method, its performance is greatly dependent on the data set for training the model. The existing data set may not cover all types of obstacles and various environmental conditions, such as different lighting and weather conditions. If the data set is to be increased, high computing resources are required. Although the application of edge computing technology can alleviate this problem to a certain extent, the dependence on data transmission speed and processing speed will also increase. Summary of the invention

[0004] In an exemplary embodiment of the present application, a radar-based traffic accident identification method and electronic device are provided to improve the identification accuracy and speed of accidents on traffic roads.

[0005] According to a first aspect of an exemplary embodiment, a method for identifying a traffic accident based on radar is provided, comprising:

[0006] Acquire traffic data and detection parameters from a radar set at a target intersection; wherein the traffic data includes a reference position and speed of each object detected by the radar, and the detection parameters include a boundary value of a detection area of ​​the radar;

[0007] Generate N trajectory maps based on traffic data and detection parameters;

[0008] Identify M trajectory images to determine whether there is a traffic accident at the target intersection; wherein the M trajectory images are part or all of the N trajectory images; M and N are integers;

[0009] If there is a traffic accident at the target intersection, the location of the accident point and the detour area caused by the traffic accident are determined.

[0010] According to a second aspect of the exemplary implementation, there is provided an electronic device, including a processor and a memory;

[0011] The memory is configured to execute:

[0012] storing traffic data and detection parameters from radars installed at target intersections;

[0013] The processor is configured to execute:

[0014] Acquire traffic data and detection parameters; wherein the traffic data includes the reference position and speed of each object detected by the radar, and the detection parameters include the boundary value of the detection area of ​​the radar;

[0015] Generate N trajectory maps based on traffic data and detection parameters;

[0016] Identify M trajectory images to determine whether there is a traffic accident at the target intersection; wherein the M trajectory images are part or all of the N trajectory images; M and N are integers;

[0017] If there is a traffic accident at the target intersection, the location of the accident point and the detour area caused by the traffic accident are determined.

[0018] According to a third aspect of an exemplary embodiment, there is provided a radar-based traffic accident identification device, comprising:

[0019] A data acquisition module, used to: acquire traffic data and detection parameters from a radar set at a target intersection; wherein the traffic data includes a reference position and speed of each object detected by the radar, and the detection parameters include a boundary value of a detection area of ​​the radar;

[0020] The trajectory map generation module is used to: generate N trajectory maps based on traffic data and detection parameters;

[0021] The data processing module is used to: identify M trajectory images to determine whether there is a traffic accident at the target intersection; wherein the M trajectory images are part or all of the N trajectory images; M and N are integers;

[0022] The data processing module is also used to: if there is a traffic accident at the target intersection, determine the location of the accident point and the detour area caused by the traffic accident.

[0023] According to a fourth aspect of an exemplary implementation, a computer storage medium is provided, in which computer program instructions are stored. When the instructions are executed on a computer, the computer executes the radar-based traffic road obstacle identification method as described in the first aspect.

[0024] In the embodiment of the present application, the traffic data and detection parameters of the radar set at the target intersection can be used. Since the traffic data includes the reference position and speed of each object detected by the radar, and the detection parameters include the boundary value of the detection area of ​​the radar, N trajectory maps can be generated based on the traffic data and detection parameters. In this way, part or all of the N trajectory maps can be selected for identification, and it can be determined whether there is a traffic accident at the target intersection. If there is a traffic accident at the target intersection, not only the location of the accident point of the traffic accident can be determined, but also the detour area caused by the traffic accident can be determined. Compared with the application of video analysis technology to identify traffic road obstacles in related technologies, it is not affected by rainy and snowy weather and light, and has a higher detection accuracy. In addition, there is no need to rely on a large amount of training data sets, which improves the recognition speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0026] Figure 1 A schematic diagram of a video screenshot of an intersection accident provided by an embodiment of the present application is exemplarily shown;

[0027] Figure 2 A flowchart of a radar-based traffic accident identification method provided in an embodiment of the present application is exemplarily shown;

[0028] Figure 3 A flow chart of a method for generating a trajectory map provided by an embodiment of the present application is exemplarily shown;

[0029] Figure 4 A schematic diagram of a trajectory diagram provided by an embodiment of the present application is exemplarily shown;

[0030] Figure 5 A flowchart of a method for determining whether a traffic accident exists at a target intersection provided by an embodiment of the present application is exemplarily shown;

[0031] Figure 6 A method flow chart of a model training process of a target neural network provided by an embodiment of the present application is exemplarily shown;

[0032] Figure 7 A schematic diagram of labeling a trajectory map provided by an embodiment of the present application is exemplarily shown;

[0033] Figure 8 A schematic diagram of a cell division provided by an embodiment of the present application is exemplarily shown;

[0034] Fig. 9 The schematic diagram of the structure of a radar-based traffic accident identification device provided in an embodiment of the present application is exemplarily shown;

[0035] Fig.10 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application is exemplarily shown. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0037] For ease of understanding, the terms involved in the embodiments of the present application are explained below:

[0038] (1) Edge computing: refers to the use of an open platform that integrates network, computing, storage, and application core capabilities on the side close to the source of objects or data to provide the nearest service. Its application is initiated on the edge side, resulting in a faster network service response, meeting the basic needs of the industry in real-time business, application intelligence, security and privacy protection, etc. Edge computing is between the physical entity and the industrial connection, or on top of the physical entity. Cloud computing can still access historical data of edge computing. The edge computing process involved in the embodiments of the present application can be performed by an edge computing device set on the roadside.

[0039] In modern urban transportation systems, video analysis technology plays a core role, especially in the detection of traffic road obstacles. Video analysis technology effectively identifies and handles various road obstacles by capturing real-time images and applying image processing and machine learning techniques. Next, several key technologies and methods for using video analysis technology for traffic obstacle detection, as well as the challenges they face in practical applications, are explained.

[0040] Among them, image feature method and behavior feature method are two main video analysis techniques. Image feature method relies on the visual features of traffic obstacles extracted from the video. Using deep learning algorithms, such as convolutional neural networks, the characteristics of accident vehicles, construction vehicles, safety fences and other obstacles can be effectively identified. Behavioral feature law analyzes dynamic behavior patterns in the video, such as vehicle behavior and pedestrian dynamics. For example, in the case of a traffic accident, people may appear in the video around the accident vehicle; in road construction areas, the activities of construction vehicles and personnel will also show specific behavior patterns. By analyzing these behaviors, the system can determine whether there are traffic obstacles in a certain area.

[0041] Although video analysis technology has shown great potential in traffic obstacle detection, it still faces some limitations and challenges in practical applications. General surveillance cameras can only cover a limited area, and it is difficult for the system to effectively detect accidents or obstacles beyond the camera's field of view. This situation is particularly prominent on highways or in more remote areas. Whether it is image feature method or behavior feature method, its performance depends greatly on the data set used to train the model. Existing data sets may not cover all types of obstacles and various environmental conditions, such as different lighting and weather conditions. This leads to the possible performance degradation of the model when facing unseen scenes. In addition, traffic obstacle detection requires the system to be able to process and analyze a large amount of video data in real time, which places high demands on computing resources. Although the application of edge computing technology can partially alleviate this problem, there is still room for improvement in data transmission and processing speed. In complex traffic environments, such as interlaced vehicles, pedestrians, bicycles, etc., it becomes more difficult to correctly identify and classify various obstacles. In addition, video quality is greatly affected by weather. For example, the accuracy of image recognition in bad weather conditions such as rain and fog will drop significantly.

[0042] To this end, an embodiment of the present application provides a radar-based traffic accident identification method, in which traffic data and detection parameters are obtained by a radar set at a target intersection, where the traffic data includes the reference position and speed of each object detected by the radar, and the detection parameters include the boundary value of the radar's detection area. N trajectory graphs are then generated based on the traffic data and the detection parameters. When the M trajectory graphs are identified, it is not only possible to determine whether there is a traffic accident at the target intersection, but also when there is a traffic accident at the target intersection, the location of the accident point of the traffic accident and the detour area caused by the traffic accident can be determined. Compared with the video analysis technology in the related art, it is not affected by the weather environment, and has lower requirements for the data set of the training model, does not require high hardware resources, and has more accurate recognition results and faster recognition speed.

[0043] For example, compared with the prior art which usually can only detect 60 meters, in the embodiment of the present application, a higher detection range can be achieved, for example, up to 300 meters, and the detection accuracy is also improved, and is not affected by rain, snow or light. It can identify many scenarios such as construction vehicles, single-vehicle accidents, abnormal parking, etc.

[0044] After introducing the design ideas of the embodiments of the present application, the following briefly introduces the application scenarios to which the technical solutions of the embodiments of the present application can be applied. It should be noted that the application scenarios introduced below are only used to illustrate the embodiments of the present application and are not limited. In specific implementation, the technical solutions provided by the embodiments of the present application can be flexibly applied according to actual needs.

[0045] refer to Figure 1, shows a schematic diagram of a video screenshot of an intersection accident, wherein: Figure 1 What is shown in the figure is a screenshot of a video taken by a monitoring device at an intersection when an actual traffic accident occurs. Traffic accidents can hardly be identified by video analysis technology. Therefore, in the embodiment of the present application, the trajectory of a traffic object (hereinafter referred to as an object, such as a vehicle) is detected based on a radar (such as a millimeter wave radar) to identify traffic obstacles; according to the movement trajectory of the object detected by the radar, a deep learning method is used to determine obstacles on the traffic road (usually caused by traffic accidents, such as the obstacle is an accident vehicle).

[0046] To further illustrate the technical solution provided by the embodiment of the present application, this is described in detail below in conjunction with the accompanying drawings and specific implementation methods. Although the embodiment of the present application provides the method operation steps shown in the following embodiments or drawings, more or fewer operation steps may be included in the method based on routine or no creative labor. In the steps where there is no necessary causal relationship logically, the execution order of these steps is not limited to the execution order provided in the embodiment of the present application.

[0047] Combine the following Figure 1 The application scenario shown is shown in Figure 2 A flow chart of a method for identifying a traffic accident based on radar is shown, and the method can be executed by an electronic device, and the electronic device can be an edge computing device. The technical solution provided in the embodiment of the present application is described.

[0048] S201: Acquire traffic data and detection parameters from a radar installed at a target intersection.

[0049] S202: Generate N trajectory maps based on traffic data and detection parameters.

[0050] S203: Identify the M trajectory images to determine whether there is a traffic accident at the target intersection.

[0051] Among them, the M trajectory images are part or all of the N trajectory images; M and N are integers.

[0052] S204: If a traffic accident occurs at the target intersection, determine the location of the accident point and the detour area caused by the traffic accident.

[0053] In the embodiment of the present application, the traffic data and detection parameters of the radar set at the target intersection can be used. Since the traffic data includes the reference position and speed of each object detected by the radar, and the detection parameters include the boundary value of the detection area of ​​the radar, N trajectory maps can be generated based on the traffic data and detection parameters. In this way, part or all of the N trajectory maps can be selected for identification, and it can be determined whether there is a traffic accident at the target intersection. If there is a traffic accident at the target intersection, not only the location of the accident point of the traffic accident can be determined, but also the detour area caused by the traffic accident can be determined. Compared with the application of video analysis technology to identify traffic road obstacles in related technologies, it is not affected by rainy and snowy weather and light, and has a higher detection accuracy. In addition, there is no need to rely on a large amount of training data sets, which improves the recognition speed.

[0054] In S201, the target intersection may be an intersection, and a radar may be set at the target intersection. The electronic device obtains traffic data and detection parameters from the radar, where the traffic data includes the reference position and speed of each object detected by the radar, and the reference position is relative to the position of the radar; the detection parameter is the boundary value of the detection area of ​​the radar. Exemplarily, the boundary value of the detection area includes the minimum lateral boundary value (X min ), maximum lateral boundary value (X max ), minimum vertical boundary value (Y min ) and the maximum longitudinal boundary value (Y max ).

[0055] In actual application, the boundary value of the radar detection area can be determined through steps A1-A3:

[0056] A1: Get the self-learning time threshold.

[0057] The self-learning time threshold can be preset and represented by T_learntime.

[0058] A2: Within the self-learning time threshold, periodically collect the first traffic data of the radar.

[0059] During the T_learntime, each frame of radar data is collected, wherein each frame of radar data is called first traffic data, including a reference position (horizontal coordinate and vertical coordinate) and a speed of each object.

[0060] A3: Determine a boundary value of a detection area of ​​the radar according to a plurality of first traffic data collected in a plurality of periods.

[0061] In this way, the boundary value of the detection area of ​​the radar can be determined according to multiple first traffic data within the T_learntime time, that is, multiple sets of coordinates. Among them, the boundary value Xmin of the detection area is the horizontal coordinate of the object with the smallest horizontal coordinate in the first traffic data; the boundary value Xmax of the detection area is the horizontal coordinate of the object with the largest horizontal coordinate in the first traffic data; the boundary value Ymin of the detection area is the vertical coordinate of the object with the smallest vertical coordinate in the first traffic data; the boundary value Ymax of the detection area is the vertical coordinate of the object with the largest vertical coordinate in the first traffic data.

[0062] In S202, when the actual obstacle recognition is performed, after the traffic data and the detection parameters are obtained, N trajectory maps can be generated based on the traffic data and the detection parameters. Figure 3 Steps S301-S303 in the figure illustrate the process of generating the trajectory map.

[0063] S301: If there are n groups of traffic data, then a first number of traffic data are selected from the n groups of traffic data.

[0064] Among them, the interval between the collection moments of two adjacent groups of traffic data is the set interval. The first number here can be represented by T_num, that is, the first T_num group can be selected from n groups of traffic data to calculate the first trajectory map. Then continue to use the next T_num group to calculate the second trajectory map, and so on. In addition, since traffic data is collected periodically, the first time length can also be used to replace the first number to screen traffic data. For example, the first time length is T_time, that is, the traffic data included in the first time length is selected to calculate the first trajectory map. The values ​​of the first time length and the first data can be set by the electronic device according to the performance of the radar and user needs.

[0065] S302: In a first amount of traffic data, for each set of traffic data, for each object in at least one object indicated by the traffic data, calculate a position of the object in a trajectory map according to a reference position and a detection parameter of the object.

[0066] In each collection cycle, a set of traffic data is collected, which can also be called a frame of traffic data. Each frame of traffic data may correspond to one or more objects, where the objects mainly refer to vehicles. Therefore, the position of the object in the trajectory map can be calculated based on the reference position and detection parameters in the traffic data.

[0067] Exemplarily, the size of the trajectory graph is determined according to the boundary value of the detection area and the set constant. For example, in order to unify the size and facilitate the representation, a set constant can be selected to scale the size of the detection area determined by the boundary value of the detection area. Among them, the set constant can be 50 and 250, so the size of the trajectory graph is as follows: (x_max-x_min) / 50*(y_max-y_min) / 250.

[0068] Optionally, the position of the object in the trajectory map is determined through steps B1-B2.

[0069] B1: Determine the lateral position of the object in the trajectory diagram according to the lateral position in the reference position of the object, the minimum lateral boundary value and the set constant.

[0070] Optionally, the lateral position of the object in the trajectory map is determined by the following formula:

[0071] Img_x = int((obj_x-x_min) / 50)

[0072] Among them, Img_x is the lateral position of the object in the trajectory map, int is the rounding operation character, obj_x is the lateral position of the reference position of the object obtained by the radar, and x_min is the minimum lateral boundary value among the boundary values ​​of the radar's detection area.

[0073] B2: Determine the longitudinal position of the object in the trajectory diagram according to the longitudinal position in the reference position of the object, the minimum longitudinal boundary value and the set constant.

[0074] Optionally, the longitudinal position of the object in the trajectory graph is determined by the following formula:

[0075] Img_y = int((obj_y-y_min) / 50)

[0076] Among them, Img_y is the longitudinal position of the object in the trajectory map, int is the rounding operation character, obj_y is the longitudinal position of the reference position of the object obtained by the radar, and y_min is the minimum longitudinal boundary value among the boundary values ​​of the radar's detection area.

[0077] S303: Determine the positions of the objects indicated by the groups of traffic data in the first amount of traffic data in the trajectory map to form a trajectory map.

[0078] Among them, the positions of multiple objects indicated by multiple sets of traffic data in the trajectory map are sequentially connected to form a trajectory map. In addition, the speed of the object in each set of traffic data can be superimposed, and the superimposed speed can be positively correlated or negatively correlated with the pixel value of the object in the corresponding trajectory map, that is, the greater the superimposed speed, the darker or brighter the pixel point of the object in the corresponding trajectory map. In the actual process, it can be determined as a positive correlation, that is, the brighter the pixel point of the object in the corresponding trajectory map.

[0079] Figure 4 A schematic diagram of a trajectory diagram provided in an embodiment of the present application, through Figure 4 It can be seen that the brighter the trajectory point, the faster the object accumulates.

[0080] In S203, after obtaining N trajectory images, some or all of them can be selected for identification, thereby determining whether there is a traffic accident at the target intersection. For example, M of them can be selected for identification. This process can be performed by Figure 5 The steps in are implemented.

[0081] S501: Input M trajectory graphs into the target neural network model.

[0082] The target neural network model includes a first CNN structure, an LSTM structure, and a second CNN structure. In the actual application process, in order to improve the recognition accuracy by ensuring the consistency of the size of the trajectory map, the M trajectory maps can be processed in the following steps C1-C1 before being input into the target neural network model:

[0083] C1: Process the M trajectory images into a first set size.

[0084] The first set size may be 360*120 (piex), so in order to ensure the consistency of the size of each trajectory graph, each trajectory graph is processed into the first set size. The specific processing method can be referred to the relevant technology, which is not limited here.

[0085] C1: According to the first set size, the M trajectory images are divided into cells of a second set size.

[0086] The second set size may be a cell of size S*S, where S may be 5. In addition, each cell outputs a set of detection results, and the detection results include whether the accident point and the detour area caused by the traffic accident are included.

[0087] S502: Identify the output result of the LSTM structure and determine whether there is a traffic accident at the target intersection.

[0088] Among them, the output result of the LSTM structure in the target neural network can be detected, and the result includes whether there is a traffic accident at the target intersection and whether there is no traffic accident at the target intersection.

[0089] In S204, when a traffic accident is detected at the target intersection, by identifying the output result of the second CNN structure, not only the location of the accident point of the traffic accident can be determined, but also the detour area caused by the traffic accident can be determined. In this process, the input of the second CNN structure is the trajectory map with the largest timestamp among the M trajectory maps, which can ensure that the recognition result is closer to the actual situation.

[0090] The above embodiment is the reasoning process of the target neural network, that is, the process of identifying the accident point and the detour area caused by the traffic accident. Next, the training process of the target neural network is described:

[0091] Figure 6 A method flow chart of a target neural network model training process provided in an embodiment of the present application, Figure 6 At least the following steps are included:

[0092] S601: Obtain training samples.

[0093] The training samples include multiple sample trajectory maps determined by traffic data obtained by radar.

[0094] S602: For each sample trajectory map, process the sample trajectory map into a first set size, and divide the trajectory map into cells of a second set size according to the first set size.

[0095] The implementation process of this step can refer to steps B11-B12, which will not be repeated here.

[0096] S603: For each cell into which the sample trajectory graph is divided, the cell where the accident point is located is marked with an inner frame, and the detour area caused by the traffic accident is marked with an outer frame.

[0097] Figure 7 A schematic diagram for marking a trajectory map provided in an embodiment of the present application, Figure 7 Includes inner frame and outer frame. Figure 8 A schematic diagram of cell division provided in an embodiment of the present application.

[0098] Among them, each cell outputs five blocks, including the confidence of the inner and outer boxes, the position (horizontal position and vertical position) and size (width and height) of the predicted box. Therefore, the block of the output layer of the entire network is (5*5*5).

[0099] S604: Input each labeled sample trajectory map into a pre-built neural network model.

[0100] S605: Determine, based on the output result of the target cell, that the loss value of the neural network model is less than a set threshold, and obtain the target neural network model.

[0101] The target cell includes the center point of the marked inner frame or the center point of the marked outer frame, and the output result includes the confidence, position and size of the corresponding frame. The output result of the non-target cell can be represented by zero. During the training process, when the loss value of the neural network model determined according to the output result of the target cell is less than the set threshold, the target neural network model can be obtained.

[0102] In a specific example, the overall loss function of the neural network model is as follows:

[0103]

[0104] Among them, λ coord Used to adjust the weight of the positioning box, λ noobj Used to shield the background. s is the horizontal position of the annotation in the s-th trajectory image in the training sample, is the horizontal position of the prediction box; y s is the horizontal position of the annotation in the s-th trajectory image in the training sample, is the horizontal position of the prediction box; w s is the horizontal position of the annotation in the s-th trajectory image in the training sample, is the horizontal position of the prediction box; h s is the horizontal position of the annotation in the s-th trajectory image in the training sample, is the horizontal position of the prediction box. In addition, c s is the confidence of the inner and outer boxes in the s-th trajectory image in the training sample, is the confidence of the predicted box. In addition, Table 1 shows the structure of a neural network.

[0105] Table 1 Structure of the neural network used in the embodiments of this application

[0106]

[0107]

[0108] In the embodiment of the present application, the edge computing device is used to analyze and process the traffic data from the radar, and a deep neural network (convolutional neural network and LSTM) is used to identify whether there are traffic obstacles on the current road that block the passage of vehicles, and then determine the location of the accident point and the detour area caused by the traffic accident. In the specific process, the traffic data of a frame of radar is first collected, and then a trajectory map is generated, and the trajectory map is pre-processed and cached. Then the trajectory map and the previous multiple trajectory maps are input into the first CNN network for feature extraction, and then the extracted features are input into the LSTM network to identify traffic obstacles. Determine whether there is a traffic obstacle on the current road. If the LSTM network determines that there is a traffic accident in the current network, the last trajectory map is input into the second CNN network used for positioning to locate the traffic accident point, and finally the actual accident location and the detour area caused by the traffic accident are output according to the accident point positioning.

[0109] like Fig. 9 As shown, based on the same inventive concept, an embodiment of the present application provides a radar-based traffic accident identification device, including a data acquisition module 91, a trajectory map generation module 92 and a data processing module 93.

[0110] The data acquisition module 91 is used to: acquire traffic data and detection parameters from a radar set at a target intersection; wherein the traffic data includes a reference position and speed of each object detected by the radar, and the detection parameters include a boundary value of a detection area of ​​the radar;

[0111] The trajectory map generation module 92 is used to: generate N trajectory maps based on traffic data and detection parameters;

[0112] The data processing module 93 is used to: identify the M trajectory images to determine whether there is a traffic accident at the target intersection; wherein the M trajectory images are part or all of the N trajectory images; M and N are integers;

[0113] The data processing module 93 is also used to determine the location of the accident point and the detour area caused by the traffic accident if there is a traffic accident at the target intersection.

[0114] In an optional implementation, the trajectory diagram generating module 92 is specifically used for:

[0115] If there are n groups of traffic data, then a first number of traffic data are selected from the n groups of traffic data; wherein the interval between the collection times of two adjacent groups of traffic data is a set interval;

[0116] In the first amount of traffic data, for each set of traffic data, for each object in at least one object indicated by the traffic data, calculating a position of the object in the trajectory map based on a reference position and a detection parameter of the object;

[0117] Determine the position of each object indicated by each group of traffic data in the first amount of traffic data in the trajectory map to form a trajectory map; wherein the pixel value of each object in the corresponding trajectory map is generated according to the speed of the object in each group of traffic data.

[0118] In an optional embodiment, the boundary values ​​of the detection area include a minimum lateral boundary value, a maximum lateral boundary value, a minimum longitudinal boundary value, and a maximum longitudinal boundary value; the size of the trajectory graph is determined according to the boundary values ​​of the detection area and a set constant;

[0119] The trajectory diagram generating module 92 is specifically used for:

[0120] Determine the lateral position of the object in the trajectory map according to the lateral position in the reference position of the object, the minimum lateral boundary value and the set constant;

[0121] The longitudinal position of the object in the trajectory map is determined according to the longitudinal position in the reference position of the object, the minimum longitudinal boundary value and the set constant.

[0122] In an optional implementation, the data processing module 93 is specifically used for:

[0123] Inputting M trajectory images into a target neural network model; wherein the target neural network model includes a first CNN structure, an LSTM structure, and a second CNN structure;

[0124] Identify the output of the LSTM structure and determine whether there is a traffic accident at the target intersection.

[0125] In an optional implementation, the data processing module 93 is specifically used for:

[0126] Identify the output result of the second CNN structure, determine the location of the accident point of the traffic accident, and the detour area caused by the traffic accident; wherein the input of the second CNN structure is a trajectory map with the largest timestamp among the M trajectory maps.

[0127] In an optional implementation, the data processing module 93 is specifically configured to, before inputting the M trajectory images into the target neural network model:

[0128] Processing the M trajectory images into a first set size;

[0129] According to the first set size, the M trajectory maps are divided into cells of a second set size respectively; wherein each cell outputs a set of detection results, and the detection results include whether an accident point and a detour area caused by the traffic accident are included.

[0130] In an optional implementation, the data processing module 93 is specifically used to train and obtain the target neural network model in the following manner:

[0131] Obtaining training samples; wherein the training samples include a plurality of sample trajectory maps determined by traffic data obtained by radar;

[0132] For each sample trajectory map, the sample trajectory map is processed into a first set size, and the trajectory map is divided into cells of a second set size according to the first set size;

[0133] For each cell divided into the sample trajectory map, the cell where the accident point is located is marked with an inner frame, and the detour area caused by the traffic accident is marked with an outer frame;

[0134] Input each labeled sample trajectory map into the pre-built neural network model;

[0135] According to the output result of the target cell, it is determined that the loss value of the neural network model is less than the set threshold, and the target neural network model is obtained; wherein the target cell includes the center point of the marked inner frame or the center point of the marked outer frame, and the output result includes the confidence, position and size of the corresponding frame.

[0136] In an optional implementation, the data acquisition module 91 is specifically used for:

[0137] Get the self-learning time threshold;

[0138] Within a self-learning time threshold, periodically collecting first traffic data of the radar;

[0139] The boundary value of the detection area of ​​the radar is determined according to a plurality of first traffic data collected in a plurality of periods.

[0140] Since the device is the device in the method in the embodiment of the present application, and the principle of solving the problem by the device is similar to that of the method, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0141] like Fig.10 As shown, based on the same inventive concept, an embodiment of the present application provides an electronic device, including a processor 101 and a memory 102 .

[0142] The memory 102 is configured to execute:

[0143] storing traffic data and detection parameters from radars installed at target intersections;

[0144] The processor 101 is configured to execute:

[0145] Acquire traffic data and detection parameters; wherein the traffic data includes the reference position and speed of each object detected by the radar, and the detection parameters include the boundary value of the detection area of ​​the radar;

[0146] Generate N trajectory maps based on traffic data and detection parameters;

[0147] Identify M trajectory images to determine whether there is a traffic accident at the target intersection; wherein the M trajectory images are part or all of the N trajectory images; M and N are integers;

[0148] If there is a traffic accident at the target intersection, the location of the accident point and the detour area caused by the traffic accident are determined.

[0149] In an optional implementation, the processor 101 is specifically configured to execute:

[0150] If there are n groups of traffic data, then a first number of traffic data are selected from the n groups of traffic data; wherein the interval between the collection times of two adjacent groups of traffic data is a set interval;

[0151] In the first amount of traffic data, for each set of traffic data, for each object in at least one object indicated by the traffic data, calculating a position of the object in the trajectory map based on a reference position and a detection parameter of the object;

[0152] Determine the position of each object indicated by each group of traffic data in the first amount of traffic data in the trajectory map to form a trajectory map; wherein the pixel value of each object in the corresponding trajectory map is generated according to the speed of the object in each group of traffic data:

[0153] In an optional embodiment, the boundary values ​​of the detection area include a minimum lateral boundary value, a maximum lateral boundary value, a minimum longitudinal boundary value, and a maximum longitudinal boundary value; the size of the trajectory graph is determined according to the boundary values ​​of the detection area and a set constant;

[0154] The processor 101 is specifically configured to execute:

[0155] Determine the lateral position of the object in the trajectory map according to the lateral position in the reference position of the object, the minimum lateral boundary value and the set constant;

[0156] The longitudinal position of the object in the trajectory map is determined according to the longitudinal position in the reference position of the object, the minimum longitudinal boundary value and the set constant.

[0157] In an optional implementation, the data processing module 93 is specifically used for:

[0158] Inputting M trajectory images into a target neural network model; wherein the target neural network model includes a first CNN structure, an LSTM structure, and a second CNN structure;

[0159] Identify the output of the LSTM structure and determine whether there is a traffic accident at the target intersection.

[0160] In an optional implementation, the processor 101 is specifically configured to execute:

[0161] Identify the output result of the second CNN structure, determine the location of the accident point of the traffic accident, and the detour area caused by the traffic accident; wherein the input of the second CNN structure is a trajectory map with the largest timestamp among the M trajectory maps.

[0162] In an optional implementation, the processor 101 is specifically configured to execute, before inputting the M trajectory images into the target neural network model:

[0163] Processing the M trajectory images into a first set size;

[0164] According to the first set size, the M trajectory maps are divided into cells of a second set size respectively; wherein each cell outputs a set of detection results, and the detection results include whether an accident point and a detour area caused by the traffic accident are included.

[0165] In an optional implementation, the processor 101 is specifically configured to train the target neural network model in the following manner:

[0166] Obtaining training samples; wherein the training samples include a plurality of sample trajectory maps determined by traffic data obtained by radar;

[0167] For each sample trajectory map, the sample trajectory map is processed into a first set size, and the trajectory map is divided into cells of a second set size according to the first set size;

[0168] For each cell divided into the sample trajectory map, the cell where the accident point is located is marked with an inner frame, and the detour area caused by the traffic accident is marked with an outer frame;

[0169] Input each labeled sample trajectory map into the pre-built neural network model;

[0170] According to the output result of the target cell, it is determined that the loss value of the neural network model is less than the set threshold, and the target neural network model is obtained; wherein the target cell includes the center point of the marked inner frame or the center point of the marked outer frame, and the output result includes the confidence, position and size of the corresponding frame.

[0171] In an optional implementation, the processor 101 is specifically configured to execute:

[0172] Get the self-learning time threshold;

[0173] Within a self-learning time threshold, periodically collecting first traffic data of the radar;

[0174] The boundary value of the detection area of ​​the radar is determined according to a plurality of first traffic data collected in a plurality of periods.

[0175] The embodiment of the present application also provides a computer storage medium, in which computer program instructions are stored. When the instructions are executed on a computer, the computer executes the steps of the above-mentioned radar-based traffic road obstacle identification method.

[0176] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0177] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0178] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0179] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0180] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for identifying traffic accidents based on radar, characterized in that: include: Acquire traffic data and detection parameters from a radar disposed at a target intersection; wherein the traffic data includes a reference position and speed of each object detected by the radar, and the detection parameters include a boundary value of a detection area of ​​the radar; Based on the traffic data and the detection parameters, generate N trajectory maps; Identify M trajectory images to determine whether there is a traffic accident at the target intersection; wherein the M trajectory images are part or all of the N trajectory images; M and N are integers; If a traffic accident occurs at the target intersection, the location of the accident point and the detour area caused by the traffic accident are determined.

2. The method according to claim 1, characterized in that The process of generating a trajectory map based on the traffic data and the detection parameters includes: If the traffic data is n groups, then a first number of traffic data is selected from the n groups of traffic data; wherein the interval between the collection times of two adjacent groups of traffic data is a set interval; In the first amount of traffic data, for each set of traffic data, for each object among at least one object indicated by the traffic data, calculating a position of the object in the trajectory map according to a reference position of the object and the detection parameter; Determine the position of each object indicated by each group of traffic data in the first number of traffic data in the trajectory map to form a trajectory map; wherein the pixel value of each object in the corresponding trajectory map is generated according to the speed of the object in each group of traffic data.

3. The method according to claim 2, characterized in that The boundary values ​​of the detection area include a minimum horizontal boundary value, a maximum horizontal boundary value, a minimum vertical boundary value and a maximum vertical boundary value; the size of the trajectory map is determined according to the boundary values ​​of the detection area and a set constant; The step of calculating the position of the object in the trajectory map according to the reference position of the object and the detection parameter comprises: Determining the lateral position of the object in the trajectory map according to the lateral position in the reference position of the object, the minimum lateral boundary value and the set constant; The longitudinal position of the object in the trajectory map is determined according to the longitudinal position in the reference position of the object, the minimum longitudinal boundary value and the set constant.

4. The method according to claim 1, characterized in that: The identifying of the M trajectory images to determine whether there is a traffic accident at the target intersection includes: Inputting the M trajectory graphs into a target neural network model; wherein the target neural network model includes a first CNN structure, an LSTM structure, and a second CNN structure; The output result of the LSTM structure is identified to determine whether there is a traffic accident at the target intersection.

5. The method according to claim 4, characterized in that The determining the location of the accident point of the traffic accident and the detour area caused by the traffic accident includes: Identify the output result of the second CNN structure, determine the location of the accident point of the traffic accident, and the detour area caused by the traffic accident; wherein the input of the second CNN structure is a trajectory map with the largest timestamp among the M trajectory maps.

6. The method according to claim 4 or 5, characterized in that: Before inputting the M trajectory graphs into the target neural network model, the method further includes: Processing the M trajectory images into a first set size; According to the first set size, the M trajectory maps are divided into cells of a second set size respectively; wherein each cell outputs a set of detection results, and the detection results include whether an accident point is included and a detour area caused by the traffic accident.

7. The method according to claim 4 or 5, characterized in that: The target neural network model is trained in the following way: Acquire training samples; wherein the training samples include a plurality of sample trajectory diagrams determined by traffic data acquired by radar; For each sample trajectory map, processing the sample trajectory map into a first set size, and dividing the trajectory map into cells of a second set size according to the first set size; For each cell into which the sample trajectory map is divided, the cell where the accident point is located is marked with an inner frame, and the detour area caused by the traffic accident is marked with an outer frame; Input each labeled sample trajectory map into the pre-built neural network model; According to the output result of the target cell, it is determined that the loss value of the neural network model is less than a set threshold, and the target neural network model is obtained; wherein, the target cell includes the center point of the marked inner frame or the center point of the marked outer frame, and the output result includes the confidence, position and size of the corresponding frame.

8. The method according to any one of claims 1 to 5, characterized in that: The method further comprises: Get the self-learning time threshold; Within the self-learning time threshold, periodically collecting first traffic data of the radar; The boundary value of the detection area of ​​the radar is determined according to a plurality of first traffic data collected in a plurality of periods.

9. An electronic device, characterized in that: including a processor and a memory; The memory is configured to execute: storing traffic data and detection parameters from radars installed at target intersections; The processor is configured to perform: Acquiring the traffic data and detection parameters; wherein the traffic data includes the reference position and speed of each object detected by the radar, and the detection parameters include the boundary value of the detection area of ​​the radar; Based on the traffic data and the detection parameters, generate N trajectory maps; Identify M trajectory images to determine whether there is a traffic accident at the target intersection; wherein the M trajectory images are part or all of the N trajectory images; M and N are integers; If a traffic accident occurs at the target intersection, the location of the accident point and the detour area caused by the traffic accident are determined.

10. The electronic device according to claim 9, characterized in that: The processor is specifically configured to execute: If the traffic data is n groups, then a first number of traffic data is selected from the n groups of traffic data; wherein the interval between the collection times of two adjacent groups of traffic data is a set interval; In the first amount of traffic data, for each set of traffic data, for each object among at least one object indicated by the traffic data, calculating a position of the object in the trajectory map according to a reference position of the object and the detection parameter; Determine the position of each object indicated by each group of traffic data in the first number of traffic data in the trajectory map to form a trajectory map; wherein the pixel value of each object in the corresponding trajectory map is generated according to the speed of the object in each group of traffic data.

Images