A method for identifying urban traffic risk hotspots

Abstract

The present invention discloses a method for identifying urban traffic risk hotspots, which belongs to the field of traffic management. The method includes the following steps: S1: obtaining a three-dimensional model of a target city, vehicle trajectory data, and weather data to build a city simulator; S2: identifying traffic violations based on target detection and relationship detection on the basis of the city simulator built in step S1; S3: extracting parking hotspots based on the city simulator built in step S1, and modeling environmental features to obtain traffic accident hotspots; S4: calculating traffic risk hotspots based on a weighted clustering algorithm for the traffic violations identified in step S2 and the traffic accident hotspots obtained in step S3. The city simulator is used to comprehensively simulate the traffic conditions in the city, and methods such as deep learning are used to timely and accurately discover urban traffic risk hotspot areas.

Description

Technical Field

[0001] The present invention belongs to the field of traffic management, and in particular relates to a method for identifying urban traffic risk hotspots. Background Art

[0002] As urban populations grow, the number of vehicles also increases, leading to a corresponding increase in urban traffic risks. Frequent urban traffic accidents pose a significant threat to the safety of life and property, as well as urban transportation operations. Furthermore, traffic violations on the roads disrupt traffic order, leading to more traffic accidents. To reduce traffic accidents and protect people's lives and property, city managers need to develop targeted traffic safety strategies for high-risk areas within the city.

[0003] The traditional method is to obtain traffic risk hotspots in the city through traffic police patrols, surveillance cameras, vehicle sensors and other equipment. These methods have problems such as incompleteness, delay and inaccuracy. Summary of the Invention

[0004] The purpose of the present invention is to propose a method for identifying urban traffic risk hotspots to overcome at least one of the above-mentioned defects in the prior art.

[0005] To achieve this object, the present invention adopts the following technical solutions:

[0006] The present invention provides a method for identifying urban traffic risk hotspots, comprising the following steps: S1: obtaining a three-dimensional model of a target city, vehicle trajectory data, and weather data to build a city simulator; S2: identifying traffic violations based on target detection and relationship detection on the basis of the city simulator built in step S1; S3: extracting parking hotspots on the basis of the city simulator built in step S1, and modeling environmental features to obtain traffic accident hotspots; S4: calculating traffic risk hotspots based on a weighted clustering algorithm for the traffic violations identified in step S2 and the traffic accident hotspots obtained in step S3.

[0007] Preferably, step S1 specifically includes the following steps: S11: obtaining a three-dimensional model of the target city and importing it into Unity to simulate the three-dimensional structure of the city, S12: obtaining vehicle trajectory data of the target city and importing it into Unity to simulate the driving conditions of the vehicle in the city, S13: obtaining weather data of the target city and importing it into Unity to simulate the operation of the vehicle in the city, and obtaining a city simulator.

[0008] Preferably, step S11 specifically includes the following steps: S111: downloading DEM data and satellite image data of the target area from the satellite map, S112: importing the DEM data and satellite image data into 3D visualization software to form a three-dimensional city model, and then exporting it as a three-dimensional city model in OBJ format, S113: importing the three-dimensional city model in OBJ format into Unity to simulate the three-dimensional structure of the city.

[0009] Preferably, step S12 specifically includes the following steps: S121: correcting the offset GPS trajectory data, and matching the trajectory points that are not on the road to the road using a map matching algorithm based on a hidden Markov model, S122: selecting the vehicle model library Realistic Car HD in Unity's AssetStore, and then adding a Camera as a driving recorder to each vehicle model, S123: selecting a vehicle model for each corrected trajectory data, and then converting the GPS point into Unity's world coordinates, marking two points in Unity to correspond to the GPS points, and then converting each GPS point into world coordinates in Unity, and the vehicle then drives to the coordinate point in sequence.

[0010] Preferably, step S13 specifically includes the following steps: S131: obtaining weather data for the period during which the city simulator is running from the weather query software, S132: obtaining Enviro 3 - Sky and Weather weather resources from Unity's AssetStore to simulate weather conditions, and S133: using Unity's weather resources to dynamically simulate the city's weather conditions according to the running time of the city simulator.

[0011] Preferably, step S2 specifically includes the following steps: S21: using a target detection model to identify driving violations of motor vehicles, S22: using a relationship detection model to identify riding violations of non-motor vehicles.

[0012] Preferably, step S21 specifically includes the following steps: S211: collecting traffic sign data sets to train the yolov4 target detection model to identify traffic signs, S212: in the city simulator, each vehicle uploads a picture captured by the camera to the yolov4 model on the server every 8-15s for identification to determine whether the picture contains a traffic sign, S213: if the picture is identified as containing a traffic sign, then the vehicle's behavior is used to determine whether the vehicle has committed a driving violation.

[0013] Preferably, step S22 specifically includes the following steps: S221: select a picture of the map software and manually annotate it through an image annotation tool, and save the annotation information into an xml format file, and then synthesize all the xml format annotation files into a txt file containing the image path, the target category included in the picture, and the location coordinate information corresponding to the target. The data set is a data set for cycling violation identification that can be recognized by the relationship detection model. S222: Then divide the data set into a training set, a verification set, and a test set, and train it on the HoiTransform relationship detection model to obtain a model for cycling violation identification. S223: After obtaining the model for cycling violation identification, the model can obtain multiple triples by identifying the picture. Indicates all the relations related to cycling violations contained in the image, S224: judging whether the relation group contains cycling violations based on multiple triples.

[0014] Preferably, in step S213, when the recognized image contains a prohibition on U-turn, the vehicle's rotation can be represented in Unity as , remember the rotation before 8-15s is ,if , then it is judged that the vehicle has made a left turn. If , then it is judged that the vehicle is making a U-turn. If , then it is judged that the vehicle has turned right, and in other cases it is judged that the vehicle is going straight.

[0015] Preferably, step S3 specifically includes the following steps: S31: extracting parking spots by vehicle speed and clustering them into parking hotspots, S32: performing feature extraction on the environmental information around the parking hotspots clustered in step S31, and S33: performing GCN modeling on the environmental features extracted in step S32 to obtain traffic accident hotspots.

[0016] Preferably, step S31 specifically includes the following steps: S311: judging the speed of each vehicle in the city simulator, and if it is less than the speed threshold, judging that the vehicle is parked here, S312: after obtaining the parking points through speed judgment, performing DBSCAN clustering on these parking points to obtain parking hotspots.

[0017] Preferably, step S32 specifically includes the following steps: S321: performing Bert-based feature extraction on the location name of each parking hotspot to obtain a feature vector, S322: performing AutoEncoder-based feature extraction on the urban points of interest around each parking hotspot, and S323: performing RNN-based feature extraction on the traffic flow data of each parking hotspot.

[0018] Preferably, step S33 specifically includes the following steps: S331: For each group of parking hotspots, use the Haversine formula to calculate the distance of each parking hotspot, which is used to construct the Laplace matrix neural network training of the graph, where the graph node features are the features extracted in steps S321, S322 and S323, S332: After obtaining further features through GCN, use a fully connected neural network to identify whether the feature is a traffic accident hotspot.

[0019] Preferably, step S4 specifically includes the following steps: S41: Each traffic risk hotspot to be clustered is represented as ,in It's latitude. is longitude, is the weight of the current point. If the current point is a traffic violation, then set If the current point is an accident hotspot, set .

[0020] S42: Normalize the weights of all points to obtain the weights , then we can use the formula Calculated and The weighted Euclidean distance of It is an adjustable coefficient. S43: Clustering is then performed according to the WDBSCAN algorithm to obtain traffic risk hotspots.

[0021] The present invention also provides an urban traffic risk hotspot identification system, including: a city simulator, a wireless communication module, and a cloud server. The city simulator is connected to the cloud server through the wireless communication module. The cloud server includes a traffic violation identification module, a traffic accident hotspot identification module, and a traffic risk hotspot identification module. The traffic violation identification module receives data from the city simulator and then identifies traffic violations based on target detection and relationship detection. The traffic accident hotspot identification module receives data from the city simulator and then extracts parking hotspots, and models environmental features to obtain traffic accident hotspots. The traffic risk hotspot identification module receives data from the traffic violation identification module and the traffic accident hotspot identification module, and then calculates traffic risk hotspots based on a weighted clustering algorithm.

[0022] The beneficial effects of the present invention are:

[0023] 1. Identify traffic risk hotspots in the city by building a city simulator in Unity. Use the city simulator to comprehensively simulate the traffic conditions in the city, and use methods such as deep learning to timely and accurately discover traffic risk hotspots in the city.

[0024] 2. Developed and integrated a series of traffic risk hotspot identification methods. First, to simulate vehicle movement using trajectory data in the city simulator, we first performed map-matching trajectory correction on the GPS trajectory data. Second, we developed algorithms for identifying traffic violation hotspots and traffic accident hotspots. Third, we used the weighted clustering algorithm WDBSCAN to cluster traffic violation and accident hotspots to identify traffic risk hotspots. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of the first embodiment of the present invention.

[0026] Figure 2 Schematic diagram of trajectory correction according to the first embodiment of the present invention.

[0027] Figure 3 This is a system block diagram of the second embodiment of the present invention.

[0028] The markings in the attached figure are: 1-city simulator, 2-wireless communication module, 3-cloud server, 31-traffic violation identification module, 32-traffic accident hotspot identification module, 33-traffic risk hotspot identification module. DETAILED DESCRIPTION

[0029] The present invention will now be further described with reference to the accompanying drawings and specific embodiments.

[0030] Example 1:

[0031] like Figure 1 As shown, a method for identifying urban traffic risk hotspots provided in this embodiment includes the following steps:

[0032] S1: Obtain the three-dimensional model of the target city, vehicle trajectories, weather and other multi-dimensional heterogeneous data to build a city simulator 1.

[0033] Step S1 specifically includes the following steps:

[0034] S11: Obtain a 3D model of the target city and import it into Unity to simulate the 3D structure of the city.

[0035] Step S11 specifically includes the following steps:

[0036] S111: Download DEM data and satellite image data of the target area from Bigemap.

[0037] S112: Import the DEM data and satellite image data into ARCSCENE software to form a three-dimensional city model, and then export the three-dimensional city model into an OBJ format.

[0038] S113: Import the OBJ format city 3D model into Unity to simulate the city’s 3D structure.

[0039] S12: Obtain vehicle trajectory data of the target city. The vehicle trajectory data in this embodiment is the vehicle trajectory data of taxis, and import it into Unity to simulate the driving conditions of vehicles in the city.

[0040] Step S12 specifically includes the following steps:

[0041] S121: Correct the offset GPS trajectory data, such as Figure 2 As shown in Figure 2, trajectory points that are not on the road are matched to the road using a map matching algorithm based on the Hidden Markov Model (HMM).

[0042] S122: Select the vehicle model library Realistic Car HD in Unity's AssetStore, and then add a Camera to each vehicle model as a driving recorder.

[0043] S123: Select a vehicle model for each corrected trajectory data, then convert the GPS point into Unity's world coordinates, mark two points in Unity to correspond to the GPS points, then convert each GPS point into Unity's world coordinates, and the vehicle then drives towards the coordinate points in sequence.

[0044] S13: Obtain weather data of the target city and import it into Unity to simulate the operation of vehicles in the city, thereby obtaining a city simulator 1.

[0045] Step S13 specifically includes the following steps:

[0046] S131: Obtain weather data for the period of time in which the city simulator 1 is running from Weather Underground.

[0047] S132: Get Enviro 3 - Sky and Weather resources from Unity's AssetStore to simulate weather conditions.

[0048] S133: Dynamically simulate the weather conditions of the city using Unity's weather resources according to the running time of the city simulator 1.

[0049] S2: Based on the city simulator 1 built in step S1, traffic violations are identified based on target detection and relationship detection. Traffic violations are divided into motor vehicle driving violations and non-motor vehicle riding violations.

[0050] Step S2 specifically includes the following steps:

[0051] S21: Use target detection model to identify driving violations of motor vehicles.

[0052] If the traffic sign prohibits U-turns and the current vehicle is making a U-turn, it is determined that the vehicle has violated the traffic sign. Step S21 specifically includes the following steps:

[0053] S211: Collect a traffic sign dataset to train the yolov4 object detection model to recognize traffic signs. This embodiment uses the TT-100K traffic sign dataset.

[0054] S212: In the city simulator 1, each vehicle uploads a picture captured by the camera to the yolov4 model on the server every 10 seconds for recognition to determine whether the picture contains a traffic sign.

[0055] S213: If the image is identified as prohibiting U-turns, the vehicle's rotation is then used to determine whether the vehicle is making a U-turn. Specifically, the vehicle's rotation can be represented in Unity as , remember the rotation 10s ago is ,if , then it is judged that the vehicle has made a left turn. If , then it is judged that the vehicle is making a U-turn. If , then it is judged that the vehicle has turned right, and in other cases it is judged that the vehicle is going straight.

[0056] S22: Use the relationship detection model to identify non-motor vehicle riding violations.

[0057] Step S22 specifically includes the following steps:

[0058] S221: Select Baidu Street View images and manually annotate them using LabelMe. Save the annotation information as an XML file. Then combine all the XML annotation files into a txt file containing the image path, the target category included in the image, and the location coordinate information corresponding to the target. This dataset is a dataset for cycling violation identification that can be recognized by the relationship detection model.

[0059] S222: The dataset is then divided into a training set, a validation set, and a test set, and then trained on the HoiTransform relationship detection model to obtain a model for identifying cycling violations.

[0060] S223: After obtaining the model for identifying cycling violations, the model can obtain multiple triples by identifying the image. Represents all the relationships related to cycling violations contained in the image.

[0061] S224: Determine whether the relationship group contains illegal riding behaviors based on multiple triples. For example, Represents the relationship between people riding bicycles. If there is a triple Indicates that the person is wearing a helmet. If this relationship does not exist, it is judged that the image contains the illegal behavior of riding without a helmet. If and It means that Overloaded. The General Technical Requirements for Electric Bicycles and the Technical Requirements for Motor Vehicle Operation Safety stipulate that electric bicycles and motorcycles with a speed of less than 20 km / h and a weight of no more than 40 kg are considered non-motor vehicles.

[0062] S3: Based on the city simulator 1 built in step S1, parking hotspots are extracted, and environmental features are modeled to obtain traffic accident hotspots.

[0063] Step S3 specifically includes the following steps:

[0064] S31: Extract parking spots based on vehicle speed and cluster them into parking hotspots.

[0065] Step S31 specifically includes the following steps:

[0066] S311: Determine the speed of each vehicle in the city simulator 1. If the speed is less than the speed threshold, the speed threshold of this embodiment is , it is determined that the vehicle is parked here.

[0067] S312: After obtaining parking spots through speed judgment, DBSCAN clustering is performed on these parking spots to obtain parking hot spots.

[0068] S32: Extract features from the surrounding environmental information of the parking hot spots clustered in step S31.

[0069] Step S32 specifically includes the following steps:

[0070] S321: Perform BERT-based feature extraction on the location name of each parking hotspot to obtain a feature vector.

[0071] S322: Perform AutoEncoder-based feature extraction on urban points of interest (POI) around each parking hotspot.

[0072] S323: Perform RNN-based feature extraction on the traffic flow data of each parking hotspot.

[0073] S33: Perform GCN modeling on the environmental features extracted in step S32 to obtain traffic accident hotspots.

[0074] Step S33 specifically includes the following steps:

[0075] S331: For each group of parking hotspots, use the Haversine formula to calculate the distance of each parking hotspot, which is used to construct the Laplacian matrix neural network training of the graph, where the graph node features are the features extracted in steps S321, S322 and S323.

[0076] S332: After obtaining further features through GCN, a simple fully connected neural network is used to identify whether the feature is a traffic accident hotspot.

[0077] S4: Traffic risk hotspots are calculated using a weighted clustering algorithm based on the traffic violations identified in step S2 and the traffic accident hotspots obtained in step S3. This step clusters the traffic violations identified in step S2 and the traffic accident hotspots identified in step S3 to determine traffic risk hotspots. However, because the distribution of accident hotspots has different influencing factors on traffic risk than the distribution of driving and cycling violations, a weighted DBSCAN algorithm is used for clustering.

[0078] Step S4 specifically includes the following steps:

[0079] S41: Each traffic risk hotspot that needs to be clustered is represented as ,in It's latitude. It is longitude, is the weight of the current point. If the current point is a traffic violation, then set If the current point is an accident hotspot, set .

[0080] S42: Normalize the weights of all points to obtain the weights , then we can use the formula Calculated and The weighted Euclidean distance of It is an adjustable coefficient and is set to 2 in this embodiment.

[0081] S43: Clustering is then performed based on the WDBSCAN algorithm to obtain traffic risk hotspots.

[0082] Taking Xiamen as an example, first, a static city simulator 1 is formed by importing the three-dimensional model of Xiamen and the multi-source heterogeneous data of Xiamen into Unity, and then the taxi trajectory data is imported to simulate the vehicle driving in the city. Then, deep learning and other methods are used to identify the traffic risk hotspots in the city simulator 1, where traffic risks are divided into traffic violations and traffic accidents. The relationship detection model HoiTransform is used to identify cycling violations, and the target detection yolo is used to identify driving violations. After obtaining traffic violations, the historical traffic accident data of Xiamen is used to model the traffic accident hotspots. GCN is used to identify traffic accident hotspots, and then the weighted clustering algorithm WDBSCAN is used to cluster the accident hotspots and traffic violations to obtain traffic risk hotspots. The innovations of the present invention are as follows:

[0083] (1) The present invention identifies traffic risk hotspots in a city by building a city simulator 1 in Unity, comprehensively simulates the traffic conditions in the city through the city simulator 1, and timely and accurately discovers traffic risk hotspots in the city through methods such as deep learning.

[0084] (2) This invention develops and integrates a series of traffic risk hotspot identification methods. The first step is to perform map matching-based trajectory correction on the GPS trajectory data in order to simulate vehicle movement using trajectory data in the city simulator 1. The second step is to develop a traffic violation hotspot identification algorithm and a traffic accident hotspot identification algorithm. The third step is to cluster traffic violation and traffic accident hotspots using the weighted clustering algorithm WDBSCAN to obtain traffic risk hotspots.

[0085] Example 2:

[0086] like Figure 3 As shown, this embodiment also provides an urban traffic risk hotspot identification system, including: a city simulator 1, a wireless communication module 2, and a cloud server 3. The city simulator 1 is connected to the cloud server 3 through the wireless communication module 2. The cloud server 3 includes a traffic violation identification module 31, a traffic accident hotspot identification module 32, and a traffic risk hotspot identification module 33. The traffic violation identification module 31 receives data from the city simulator 1, and then identifies traffic violations based on target detection and relationship detection. The traffic accident hotspot identification module 32 receives data from the city simulator 1, and then extracts parking hotspots, and models environmental features to obtain traffic accident hotspots. The traffic risk hotspot identification module 33 receives data from the traffic violation identification module 31 and the traffic accident hotspot identification module 32, and then calculates traffic risk hotspots based on a weighted clustering algorithm.

[0087] This system builds a simulator to sense road information in the city through vehicles or people equipped with intelligent recognition, and transmits the sensed information to the cloud server 3 through wireless communication. The cloud server 3 then identifies and clusters this information to obtain traffic risk hotspots.

[0088] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for identifying urban traffic risk hotspots, characterized in that: The following steps are involved: S1: Obtain the 3D model of the target city, vehicle trajectory data, and weather data to build a city simulator; S2: Based on the city simulator built in step S1, traffic violations are identified based on yolov4 target detection and HoiTransform relationship detection; specifically, the following steps are included: S21: Use the yolov4 target detection model to identify driving violations of motor vehicles; specifically, the following steps are included: S211: Collect traffic sign datasets to train the yolov4 target detection model to identify traffic signs; S212: In the city simulator, each vehicle uploads a picture captured by the camera to the yolov4 target detection model on the server every 8-15 seconds to identify whether the picture contains a traffic sign; S213: If the image is recognized to contain a traffic sign, then the vehicle's behavior is used to determine whether the vehicle has committed a driving violation. This specifically includes the following steps: When the image contains a prohibited U-turn, the vehicle's rotation can be represented in Unity as (x, y, z). The rotation 8-15 seconds ago is (x ′ ,y ′ ,z ′ ), if 300 ≥ yy ′ ≥240, then it is judged that the vehicle has turned left. If 240>yy ′ ≥120, then the vehicle is judged to be making a U-turn. If 120>yy ′ If the value is ≥30, the vehicle is judged to have turned right, otherwise it is judged to have gone straight; S22: Using the HoiTransform relationship detection model to identify non-motor vehicle riding violations; specifically, the following steps are included: S221: Select an image from the map software and manually annotate it using an image annotation tool. Save the annotation information as an XML file. Then, combine all the XML annotation files into a txt file containing the image path, the object category included in the image, and the location coordinates of the object. This txt file is a dataset for cycling violation recognition that can be identified by the HoiTransform relationship detection model. S222: The dataset is further divided into a training set, a validation set, and a test set, and then trained on the HoiTransform relationship detection model to obtain a HoiTransform relationship detection model for identifying cycling violations; S223: After obtaining the HoiTransform relationship detection model for cycling violation identification, the model can obtain multiple triples by identifying the image.<object1,predicate,object2> , the triple<object1,predicate,object2> Indicates all the relationships related to cycling violations contained in the image; S224: determining whether the relationship group includes a cycling violation based on the multiple triples; S3: Extract parking hotspots based on the city simulator built in step S1, and model the environmental features to obtain traffic accident hotspots; S4: Calculate traffic risk hotspots based on the weighted clustering algorithm for the traffic violations identified in step S2 and the traffic accident hotspots obtained in step S3.

2. The urban traffic risk hotspot identification method according to claim 1, characterized in that: Step S1 specifically includes the following steps: S11: Obtain a 3D model of the target city and import it into Unity to simulate the 3D structure of the city; S12: Obtain vehicle trajectory data in the target city and import it into Unity to simulate vehicle driving in the city; S13: Obtain weather data of the target city and import it into Unity to simulate the operation of vehicles in the city to obtain a city simulator.

3. The urban traffic risk hotspot identification method according to claim 2 is characterized in that: Step S11 specifically includes the following steps: S111: Download DEM data and satellite image data of the target area from the satellite map; S112: Importing the DEM data and satellite image data into 3D visualization software to form a three-dimensional city model, and then exporting the three-dimensional city model into an OBJ format; S113: Import the OBJ format city 3D model into Unity to simulate the city’s 3D structure.

4. The urban traffic risk hotspot identification method according to claim 1 is characterized in that: Step S12 specifically includes the following steps: S121: Correcting the offset GPS trajectory data, and matching trajectory points that are not on the road to the road using a map matching algorithm based on a hidden Markov model; S122: Select the vehicle model library Realistic Car HD in Unity's AssetStore, and then add a Camera to each vehicle model as a driving recorder; S123: Select a vehicle model for each corrected trajectory data, then convert the GPS point into Unity's world coordinates, mark two points in Unity to correspond to the GPS points, then convert each GPS point into Unity's world coordinates, and the vehicle then drives towards the coordinate points in sequence.

5. The urban traffic risk hotspot identification method according to claim 2 is characterized in that: Step S13 specifically includes the following steps: S131: Obtaining weather data for the period during which the city simulator is running from weather query software; S132: Get Enviro 3-Sky and Weather resources from Unity's AssetStore to simulate weather conditions; S133: Dynamically simulate the weather conditions of the city using Unity's weather resources according to the running time of the city simulator.

6. The urban traffic risk hotspot identification method according to claim 1 is characterized in that: Step S3 specifically includes the following steps: S31: Extract parking spots based on vehicle speed and cluster them into parking hotspots; S32: extracting features from the surrounding environment information of the parking hot spots clustered in step S31; S33: Perform GCN modeling on the environmental features extracted in step S32 to obtain traffic accident hotspots.

7. The urban traffic risk hotspot identification method according to claim 6 is characterized in that: Step S31 specifically includes the following steps: S311: Determine the speed of each vehicle in the city simulator, and if the speed is less than a speed threshold, determine that the vehicle is parked here; S312: After obtaining parking spots through speed judgment, DBSCAN clustering is performed on these parking spots to obtain parking hot spots.

8. The urban traffic risk hotspot identification method according to claim 6 is characterized in that: Step S32 specifically includes the following steps: S321: Perform BERT-based feature extraction on the location name of each parking hotspot to obtain a feature vector; S322: Performing AutoEncoder-based feature extraction on urban points of interest around each parking hotspot; S323: Perform RNN-based feature extraction on the traffic flow data of each parking hotspot.

9. The urban traffic risk hotspot identification method according to claim 8 is characterized in that: Step S33 specifically includes the following steps: S331: For each group of parking hotspots, use the Haversine formula to calculate the distance of each parking hotspot, which is used to construct the Laplacian matrix neural network training of the graph, where the graph node features are the features extracted in steps S321, S322 and S323; S332: After obtaining further features through GCN, a fully connected neural network is used to identify whether the feature is a traffic accident hotspot.

10. The urban traffic risk hotspot identification method according to claim 1, characterized in that: Step S4 The specific steps include: S41: Each traffic risk hotspot that needs to be clustered is represented as p i =(x i ,y i ,c i ), where x i It is latitude, y i is longitude, c i is the weight of the current point. If the current point is a traffic violation, set c i =1, if the current point is an accident hotspot, set c i =30; S42: Normalize the weights of all points to obtain weight w i , then we can use the formula Calculate the p i and p j The weighted Euclidean distance of , where ω is an adjustable coefficient; S43: Clustering is then performed based on the WDBSCAN algorithm to obtain traffic risk hotspots.

11. Urban traffic risk hotspot identification system, characterized by: include: City simulator, wireless communication module, and cloud server; The city simulator is connected to the cloud server via a wireless communication module; The cloud server includes a traffic violation identification module, a traffic accident hotspot identification module, and a traffic risk hotspot identification module; The traffic violation identification module receives the data from the city simulator and then identifies traffic violations based on yolov4 target detection and HoiTransform relationship detection; specifically, it includes the following steps: S21: Use the yolov4 target detection model to identify driving violations of motor vehicles; specifically, the following steps are included: S211: Collect traffic sign datasets to train the yolov4 target detection model to identify traffic signs; S212: In the city simulator, each vehicle uploads a picture captured by the camera to the yolov4 target detection model on the server every 8-15 seconds to identify whether the picture contains a traffic sign; S213: If the image is recognized to contain a traffic sign, then the vehicle's behavior is used to determine whether the vehicle has committed a driving violation. This specifically includes the following steps: When the image contains a prohibited U-turn, the vehicle's rotation can be represented in Unity as (x, y, z). The rotation 8-15 seconds ago is (x ′ ,y ′ ,z ′ ), if 300 ≥ yy ′ ≥240, then it is judged that the vehicle has turned left. If 240>yy ′ ≥120, then the vehicle is judged to be making a U-turn. If 120>yy ′ If the value is ≥30, the vehicle is judged to have turned right, otherwise it is judged to have gone straight; S22: Using the HoiTransform relationship detection model to identify non-motor vehicle riding violations; specifically, the following steps are included: S221: Select an image from the map software and manually annotate it using an image annotation tool. Save the annotation information as an XML file. Then, combine all the XML annotation files into a txt file containing the image path, the object category included in the image, and the location coordinates of the object. This txt file is a dataset for cycling violation recognition that can be identified by the HoiTransform relationship detection model. S222: The dataset is further divided into a training set, a validation set, and a test set, and then trained on the HoiTransform relationship detection model to obtain a HoiTransform relationship detection model for identifying cycling violations; S223: After obtaining the HoiTransform relationship detection model for cycling violation identification, the model can obtain multiple triples by identifying the image.<object1,predicate,object2> , the triple<object1,predicate,object2> Indicates all the relationships related to cycling violations contained in the image; S224: determining whether the relationship group includes a cycling violation based on the multiple triples; The traffic accident hotspot identification module receives data from the city simulator, extracts parking hotspots, and models environmental features to obtain traffic accident hotspots; The traffic risk hotspot identification module receives data from the traffic violation identification module and the traffic accident hotspot identification module, and then calculates traffic risk hotspots based on a weighted clustering algorithm.

Images