Driving track extraction and processing method and system based on driving recorder data

Abstract

The present invention discloses a driving trajectory extraction and processing method and system based on driving recorder data, including obtaining an urban road network structure diagram of a target city, reading the GPS data of a target vehicle according to the driving recorder data, matching the GPS trajectory sequence with the urban road network structure diagram and converting it into a road network-level trajectory sequence; determining the driving scene of the target vehicle, identifying other vehicles in the driving scene for vehicle-to-vehicle relationship analysis; constructing a risk assessment model for the driving scene, assessing the risk distribution of the driving scene based on vehicle-to-vehicle relationship combined with driving comfort, predicting the trend of risk distribution changes to generate early warning information, and adaptively increasing the sampling frequency and enhanced data of driving recorder data, and updating the risk distribution of the driving scene in real time. The present invention processes driving recorder data, analyzes vehicle danger distribution according to driving trajectories, uses prediction methods to timely discover and respond to road events, and improves road safety and vehicle early warning capabilities.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle trajectory processing, and more specifically, to a vehicle trajectory extraction and processing method and system based on vehicle recorder data. Background Art

[0002] Smart driving recorders are an automotive accessories industry that has been transformed in recent years with the rapid development of the artificial intelligence industry. Through built-in cameras and sensors, it can record various situations in the vehicle's driving process in real time through video and audio. Existing smart driving recorders can also be equipped with target detection systems to detect and mark target images such as license plates and people in the picture in real time, providing evidence for traffic accidents. It can also be used as a feedback tool for driving behavior, providing users with a more convenient user experience.

[0003] The existing network-free driving recorders, although equipped with GPS modules, cannot show the driving trajectory to users. They can only add the latitude and longitude information of the positioning points in the video screen, resulting in relatively simple functions. The driving route cannot be displayed on the recorder terminal, nor can the positioning points be uploaded to the server for subsequent users to view the driving route on the mobile terminal. The intelligence of vehicles is of great significance in helping to solve road traffic safety and congestion problems. Combined with intelligent traffic control technology, the trajectories of other vehicles in the current driving scene are analyzed and extracted, and their risk distribution is managed and evaluated to reduce the probability of traffic accidents and effectively alleviate traffic congestion and other problems, so as to create a safe and comfortable transportation environment. Therefore, how to extract driving trajectories based on driving recorder data and use the distribution of regional risk for prediction is one of the problems that need to be solved. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a method and system for extracting and processing a driving trajectory based on driving recorder data.

[0005] A first aspect of the present invention provides a method for extracting and processing a driving trajectory based on driving recorder data, comprising:

[0006] Obtaining road data of the target city, preprocessing the road data to generate a city road network structure diagram, reading GPS data of the target vehicle according to the driving recorder data, matching the GPS trajectory sequence with the city road network structure diagram and converting it into a road network-level trajectory sequence;

[0007] Determine the driving scene of the target vehicle according to the trajectory sequence at the road network level, identify other vehicles in the driving scene based on the driving recorder data of the target vehicle, and perform vehicle-to-vehicle relationship analysis through the track data of other vehicles;

[0008] Constructing a risk assessment model for driving scenarios, evaluating the risk distribution of driving scenarios based on the vehicle-to-vehicle relationship and driving comfort according to the risk assessment model, predicting the risk distribution at a preset time in the future, and obtaining a changing trend of the risk distribution;

[0009] The early warning information is generated based on the risk change trend, and the sampling frequency and enhanced data of the driving recorder data are adaptively increased to update the risk distribution of the driving scene in real time.

[0010] In this solution, the GPS data of the target vehicle is read according to the data from the driving recorder, and the GPS trajectory sequence is matched with the urban road network structure diagram and converted into a road network-level trajectory sequence, specifically:

[0011] Obtaining the road network data of the target city, simplifying the road network data, retaining the road network data of the preset road grade, gridding the road network data, and guiding the merging of intersections of the same grid according to the connectivity relationship of the intersections in the road network;

[0012] The rationality of the merged road network is verified based on traffic expertise and real urban road traffic conditions. Intersections are extracted as nodes from the simplified road network data, and the road sections between intersections are used as edge structures. The edge attributes of the edge structures are set to construct the urban road network structure diagram.

[0013] Extracting the dashcam data of the vehicle in the historical time period, separating the GPS track points in the dashcam data, generating a GPS track sequence in combination with the time sequence, segmenting the GPS track sequence, and removing the track segments without movement;

[0014] Mapping the segmented trajectory segments to the urban road network structure diagram, setting a matching window to calculate the similarity between the turning point and the intersection position, and when the position similarity between the adjacent turning points and the corresponding intersection meets the preset standard, determining whether the shortest path between the intersections is greater than a preset distance threshold;

[0015] When the distance is greater than the preset distance threshold, matching points are selected in the path between the intersections to set the matching window for re-matching, traversing all the intersections in the urban road network structure diagram, and converting the GPS data into a road network-level trajectory sequence.

[0016] In this solution, the driving scene of the target vehicle is determined according to the trajectory sequence at the road network level, and other vehicles in the driving scene are identified based on the driving recorder data of the target vehicle, specifically:

[0017] Using the road network-level trajectory sequence to perform driving positioning, and determining the traffic facility resources at the positioning location in the target city using the driving recorder data and traffic monitoring data stored in the cloud according to the driving positioning;

[0018] Constructing underlying reference information of a driving scene based on the traffic facility resources, reading real-time driving recorder data of a target vehicle to generate a video frame sequence of the driving scene, and annotating the video frame sequence using the underlying reference information;

[0019] Other vehicles are identified in the video frame sequence. The Yolov3 algorithm is used as the backbone extraction network. The preprocessed video frame images are imported into the backbone extraction network to construct a feature pyramid to obtain a multi-scale feature map.

[0020] Introducing a spatial attention mechanism and a channel attention mechanism, assigning weight information to the multi-scale feature map to update the feature distribution, recalibrating the feature map, and fusing the recalibrated feature maps by aggregating context feature maps of different ranges to obtain an optimized feature expression;

[0021] The optimized feature expression and the preset texture features and shape features are fused to generate multi-dimensional fusion features in the video scene, and the anchor frame is calibrated according to the multi-dimensional fusion features to output the target information and position information of other vehicles.

[0022] In this solution, the vehicle-to-vehicle relationship analysis is performed using the track data of other vehicles, specifically:

[0023] In the driving scene, a reference coordinate system is constructed according to the underlying reference information, vertex coordinates and centroid coordinates of anchor frames of other vehicles are obtained in the reference coordinate system, and driving trajectory fitting of other vehicles is performed based on the vertex coordinates and centroid coordinates;

[0024] The centroid coordinates are used as tracking corner points, the centroid coordinates are diluted at preset time intervals, the driving trajectory is obtained and matched with the corresponding vehicle, and the driving trajectory in the current driving scene is visualized;

[0025] Acquire motion features according to the driving trajectories of the target vehicle and other vehicles, encode the relative positions of the target vehicle and other vehicles in the driving scene at different times, combine the relative position codes with the motion features, and construct a motion feature vector;

[0026] Adding a spatial position label to the motion feature vector based on the reference coordinate system, calculating a spatial attention weight for the motion feature vector with the label, extracting the attention degree of the target vehicle in space, and weighting the motion feature vector using the spatial attention weight;

[0027] The weighted motion feature vector is used to obtain interaction information of other vehicles through convolution and maximum pooling operations, and the vehicle-to-vehicle relationship with the target vehicle is determined based on the interaction information.

[0028] In this solution, a risk assessment model for driving scenarios is constructed, and the risk distribution of driving scenarios is assessed based on the vehicle-to-vehicle relationship and driving comfort according to the risk assessment model, specifically:

[0029] Acquire motion features from a trajectory sequence of a road network level in a preset historical period of the target vehicle, cluster the trajectory sequence using the motion features, and cluster different sequence segments to generate a number of sequence segment subsets;

[0030] Acquire the driving recorder data of each sequence segment in the sequence segment subset, determine the corresponding driving scene, perform redundancy removal processing on the driving scene, match the motion feature with the processed driving scene, and use the motion feature as the driving preference of each driving scene;

[0031] Retrieving the corresponding driving preference according to the current driving scene, and calculating the similarity between the motion characteristics of the current driving scene and the driving preference to obtain a driving comfort evaluation;

[0032] Analyze risk factors according to accident instances in driving scenarios corresponding to historical preset time periods, match the risk factors with vehicle-to-vehicle relationships, and construct a risk assessment model for driving scenarios;

[0033] Obtain the Euclidean distance between the target vehicle and other vehicles in the driving scene, calculate the weight information of other vehicles based on the Euclidean distance and the driving comfort of the target vehicle, and extract parameters based on the risk factors through the driving trajectories of other vehicles;

[0034] In the risk assessment model, risk analysis is performed based on risk parameters and weight information, and the risks of other vehicles are spatially differentiated in the driving scene to visualize the risk distribution.

[0035] In this solution, warning information is generated based on the risk change trend, and the sampling frequency and enhanced data of the driving recorder data are adaptively increased, specifically:

[0036] Select the risk distribution of the previous n moments in the current driving scene to construct a risk distribution vector for a preset time slice, import the risk vector into the Bi-LSTM network, and use hidden state update to obtain the risk distribution change trend;

[0037] Obtaining the risk distribution after a preset time according to the change trend, and generating warning information if the risk corresponding to other vehicles is greater than a preset threshold;

[0038] The deviation information between the risk degree and the preset threshold is obtained to evaluate the warning level of the warning information, and the warning level is weighted based on the meteorological data. The weighted warning level is used to adaptively change the sampling frequency of the driving recorder data and the enhanced data of the driving recorder screen.

[0039] The second aspect of the present invention further provides a driving trajectory extraction and processing system based on driving recorder data, the system comprising: a memory, a processor, the memory comprising a driving trajectory extraction and processing method program based on driving recorder data, the driving trajectory extraction and processing method program based on driving recorder data when executed by the processor implements the following steps:

[0040] Obtaining road data of the target city, preprocessing the road data to generate a city road network structure diagram, reading GPS data of the target vehicle according to the driving recorder data, matching the GPS trajectory sequence with the city road network structure diagram and converting it into a road network-level trajectory sequence;

[0041] Determine the driving scene of the target vehicle according to the trajectory sequence at the road network level, identify other vehicles in the driving scene based on the driving recorder data of the target vehicle, and perform vehicle-to-vehicle relationship analysis through the track data of other vehicles;

[0042] Constructing a risk assessment model for driving scenarios, evaluating the risk distribution of driving scenarios based on the vehicle-to-vehicle relationship and driving comfort according to the risk assessment model, predicting the risk distribution at a preset time in the future, and obtaining a changing trend of the risk distribution;

[0043] The early warning information is generated based on the risk change trend, and the sampling frequency and enhanced data of the driving recorder data are adaptively increased to update the risk distribution of the driving scene in real time.

[0044] The present invention discloses a driving trajectory extraction and processing method and system based on driving recorder data, including obtaining an urban road network structure diagram of a target city, reading the GPS data of a target vehicle according to the driving recorder data, matching the GPS trajectory sequence with the urban road network structure diagram and converting it into a road network-level trajectory sequence; determining the driving scene of the target vehicle, identifying other vehicles in the driving scene for vehicle-to-vehicle relationship analysis; constructing a risk assessment model for the driving scene, assessing the risk distribution of the driving scene based on vehicle-to-vehicle relationship combined with driving comfort, predicting the trend of risk distribution changes to generate early warning information, and adaptively increasing the sampling frequency and enhanced data of driving recorder data, and updating the risk distribution of the driving scene in real time. The present invention processes driving recorder data, analyzes vehicle danger distribution according to driving trajectories, uses prediction methods to timely discover and respond to road events, and improves road safety and vehicle early warning capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A flow chart showing a method for extracting and processing a driving trajectory based on driving recorder data of the present invention is shown;

[0046] Figure 2A flow chart showing the present invention's method of performing vehicle-to-vehicle relationship analysis using track data of other vehicles;

[0047] Figure 3 A flow chart showing a risk assessment model for constructing a driving scenario according to the present invention is shown;

[0048] Figure 4 The block diagram of the driving trajectory extraction and processing system based on driving recorder data of the present invention is shown. DETAILED DESCRIPTION

[0049] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0050] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0051] Figure 1 The flowchart of the driving trajectory extraction and processing method based on driving recorder data of the present invention is shown.

[0052] like Figure 1 As shown, the first aspect of the present invention provides a method for extracting and processing a driving trajectory based on driving recorder data, comprising:

[0053] S102, obtaining road data of a target city, preprocessing the road data to generate a city road network structure diagram, reading GPS data of a target vehicle according to the driving recorder data, matching the GPS trajectory sequence with the city road network structure diagram and converting it into a road network-level trajectory sequence;

[0054] S104, determining a driving scene of the target vehicle according to the trajectory sequence at the road network level, identifying other vehicles in the driving scene based on the driving recorder data of the target vehicle, and performing vehicle-to-vehicle relationship analysis through the trajectory data of other vehicles;

[0055] S106, constructing a risk assessment model for driving scenarios, evaluating the risk distribution of driving scenarios based on the vehicle-to-vehicle relationship and driving comfort according to the risk assessment model, predicting the risk distribution at a preset time in the future, and obtaining a risk distribution change trend;

[0056] S108, generating early warning information based on the risk change trend, and adaptively increasing the sampling frequency of the driving recorder data and enhancing the data, and updating the risk distribution of the driving scene in real time.

[0057] It should be noted that the road network data of the target city is obtained. The urban road network is usually composed of many different elements, including main roads, secondary roads, urban ring roads, highways, expressways, pedestrian roads, bicycle roads, etc.; the road network data is simplified, the road network data of the preset road level is retained, other types of roads are deleted, and isolated nodes that are not connected to the main road network due to road removal are deleted; the road network data is gridded, and the intersections of the same grid are guided to be merged according to the connectivity relationship of the intersections in the road network; the rationality of the merged road network is verified according to traffic professional knowledge and the actual road traffic conditions in the city, the intersections are extracted as nodes in the simplified road network data, the sections between the intersections are used as edge structures, and the edge attributes of the edge structures are set to construct the urban road network structure diagram; the driving recorder data of the vehicle in the historical time period is extracted, the GPS track points are separated from the driving recorder data, the GPS track sequence is generated in combination with the time series, and the GPS track sequence is segmented. Since some driving recorder devices will continue to upload driving tracks after the vehicle is turned off, the track segments without movement are eliminated;

[0058] The segmented trajectory segments are mapped to the urban road network structure diagram, a matching window is set to calculate the similarity between the turning point and the intersection position, and the turning point is matched to the most likely road. After the turning point is matched with the intersection, if the position similarity between the next adjacent turning point and the corresponding intersection meets the preset standard, it is determined whether the shortest path between the intersections is greater than the preset distance threshold; when it is greater than the preset distance threshold, a matching point is selected in the path between the intersections to set a matching window for re-matching. When the match is successful, it is determined that the target vehicle's driving segment is successfully matched, and all intersections in the urban road network structure diagram are traversed to convert the GPS data into a road network-level trajectory sequence.

[0059] Figure 2 A flow chart of the present invention for performing vehicle-to-vehicle relationship analysis using track data of other vehicles is shown.

[0060] According to an embodiment of the present invention, the vehicle-to-vehicle relationship analysis is performed using the track data of other vehicles, specifically:

[0061] S202, constructing a reference coordinate system in the driving scene according to the underlying reference information, obtaining vertex coordinates and centroid coordinates of anchor frames of other vehicles in the reference coordinate system, and fitting driving trajectories of other vehicles based on the vertex coordinates and centroid coordinates;

[0062] S204, using the centroid coordinates as tracking corner points, diluting the centroid coordinates at preset time intervals, obtaining a driving trajectory to match with a corresponding vehicle, and visually displaying the driving trajectory in the current driving scene;

[0063] S206, obtaining motion features according to the driving trajectories of the target vehicle and other vehicles, encoding the relative positions of the target vehicle and other vehicles in the driving scene at different times, combining the relative position codes with the motion features, and constructing a motion feature vector;

[0064] S208, adding a spatial position label to the motion feature vector based on the reference coordinate system, calculating a spatial attention weight for the labeled motion feature vector, extracting the attention degree of the target vehicle in space, and weighting the motion feature vector using the spatial attention weight;

[0065] S210, obtaining interaction information of other vehicles through convolution and maximum pooling operations on the weighted motion feature vector, and determining the vehicle-to-vehicle relationship with the target vehicle according to the interaction information.

[0066] It should be noted that the road network-level trajectory sequence is used for driving positioning, and the traffic facility resources at the positioning location are determined in the target city using the driving recorder data and traffic monitoring data stored in the cloud according to the driving positioning; the underlying reference information of the driving scene is constructed according to the traffic facility resources, the real-time driving recorder data of the target vehicle is read to generate a video frame sequence of the driving scene, and the underlying reference information is used to annotate the video frame sequence; other vehicles are identified in the video frame sequence, and the Yolov3 algorithm is used as the backbone extraction network, the preprocessed video frame image is imported into the backbone extraction network, and a feature pyramid is constructed to obtain a multi-scale feature map; Introduce spatial attention mechanism and channel attention mechanism, assign weight information to the multi-scale feature map to update feature distribution, recalibrate feature map, fuse recalibrated feature map by aggregating context feature maps of different ranges, and obtain optimized feature expression; obtain massive automobile image information, analyze shallow features of different types of vehicles through binarization and morphological processing methods to build preset texture features and shape feature libraries, fuse optimized feature expressions and preset texture features and shape features, generate multi-dimensional fusion features in video scenes, calibrate anchor frames according to the multi-dimensional fusion features, and output target information and position information of other vehicles. Obtain image data of vehicles in motion from dashcam data in the cloud, obtain labeled image data through manual annotation and experience annotation methods, establish vehicle data set based on dashcam perspective, and conduct network training. YOLO recognition method can simultaneously detect and identify the location information and category information of target objects, with extremely fast detection speed, low false detection rate, and strong generalization ability.

[0067] It should be noted that the RNN network and other recurrent neural network methods are used to extract the vehicle motion features under time series, and the vehicle motion features share a feature space. In order to reduce the spatial complexity of the motion feature extraction module and the time complexity of network training, all vehicles in the driving scene share the motion feature network parameters, and the target vehicle and other vehicles use the same network parameters. In the driving scene, other vehicles around the target vehicle will have a great impact on its motion state, so it is necessary to observe and judge the interaction between the target vehicle and other vehicles. For example, when the current vehicle slows down, the target vehicle needs to brake. When the target vehicle needs to change lanes and overtake, it is necessary to observe the traffic status of the target lane and the vehicles in front and behind. The current driving scene is gridded, and the vehicle's motion feature vector is multiplied by the position code to obtain a motion feature vector with spatial position information. Different coding information is injected into each spatial position using the reference coordinate system, so that the network can distinguish the spatial position. The input features are subjected to maximum pooling and average pooling operations to generate corresponding feature maps, which are then spliced ​​according to channels to fuse spatial features. The spliced ​​feature maps are convoluted to obtain fused features, spatial attention weights are obtained, and the motion feature vectors are weighted to extract features of the spatial position of the vehicle. The local receptive field is used to obtain the workshop relationship between vehicles, and the embedding layer is used to change the dimension of the workshop relationship vector to output the corresponding workshop relationship.

[0068] Figure 3 A flow chart of constructing a risk assessment model for a driving scenario according to the present invention is shown.

[0069] According to an embodiment of the present invention, a risk assessment model for driving scenarios is constructed, and the risk distribution of driving scenarios is assessed based on the vehicle-to-vehicle relationship and driving comfort according to the risk assessment model, specifically:

[0070] S302, acquiring motion features from a trajectory sequence of a road network level in a preset historical period of the target vehicle, clustering the trajectory sequence using the motion features, and clustering different sequence segments to generate a plurality of sequence segment subsets;

[0071] S304, acquiring the driving recorder data of each sequence segment in the sequence segment subset, determining the corresponding driving scene, performing redundancy removal processing on the driving scene, matching the motion feature with the processed driving scene, and using the motion feature as the driving preference of each driving scene;

[0072] S306, searching for a corresponding driving preference according to the current driving scene, and calculating a similarity between the motion characteristics of the current driving scene and the driving preference to obtain a driving comfort evaluation;

[0073] S308, analyzing risk factors according to accident instances in driving scenarios corresponding to historical preset time periods, matching the risk factors with vehicle-to-vehicle relationships, and constructing a risk assessment model for driving scenarios;

[0074] S310, obtaining the Euclidean distance between the target vehicle and other vehicles in the driving scene, calculating the weight information of other vehicles according to the Euclidean distance and the driving comfort of the target vehicle, and extracting parameters based on the risk factors through the driving trajectories of other vehicles;

[0075] S312, performing risk analysis based on risk parameters and weight information in the risk assessment model, and performing spatial differentiation of the risks of other vehicles in the driving scene to visualize the risk distribution.

[0076] It should be noted that relevant accident instances are extracted from the driving scenarios corresponding to the preset historical time periods, and the dominant factors of the accident instances are analyzed. The vehicle-to-vehicle relationships are analyzed among the dominant factors, including overtaking, lane changing and parallel driving. The dominant factors are subjected to principal component analysis, and a preset number of principal component factors are selected. The correlation with the driving trajectory is calculated to obtain the risk factors corresponding to different vehicle-to-vehicle relationships. An evaluation system is constructed according to the preset threshold interval, and score information of different risk factors is generated. The score information is aggregated to generate the final score, and the risk value is generated according to the weight information corresponding to other vehicles themselves.

[0077] It should be noted that the short-term prediction of driving scene risk based on historical risk distribution data uses the accumulation of risk distribution of each driving scene in the past period of time to predict the risk in the driving scene sequence in the future. Since the movement speed of the vehicle is generally fast during driving, the driving environment changes rapidly, so the set time slice and the length of the prediction time series cannot be too large. The Bi-LSTM network is used to perform short-term prediction of risk distribution, and the historical driving scene data containing risk distribution is used to construct training samples for model training. The risk distribution of the first n moments in the current driving scene is selected to construct the risk distribution vector of the preset time slice, and the risk vector is imported into the Bi-LSTM network. The risk distribution change trend is obtained by using hidden state update; the risk distribution after the preset time is obtained according to the change trend. If the risk corresponding to other vehicles is greater than the preset threshold, an early warning information is generated; the deviation information of the risk and the preset threshold is obtained to evaluate the warning level of the early warning information, and the warning level is weighted based on the meteorological data, and the weighted warning level is used to adaptively change the sampling frequency of the driving recorder data and the enhanced data of the driving recorder screen.

[0078] Obtain the risk distribution of the current driving scene, use the risk distribution as one of the optimization indicators when planning the current path of the target vehicle, and select the path points with the lowest risk in the driving scene to generate the planned path of the target vehicle in series. When driving in bad weather such as rain and snow, in addition to the risk of sliding due to low ground resistance, high-intensity reflections can also lead to problems such as inability to dodge. In the data captured by the dash cam, the edge features between the vehicle and the background are weakened, and the detection accuracy is low. Therefore, the adaptive stretching histogram method is used to improve the contrast between the vehicle and the ground and the sky, reduce the missed detection rate of other vehicles, and increase the confidence in the identification of other vehicles, which helps to better detect targets.

[0079] The road network-level trajectory sequence of the target vehicle within the historical preset time period is divided according to the travel period, and the driving trajectory heat map of different travel periods is generated based on the urban road network structure diagram of the target city. The driving trajectory heat map of the same travel period is subjected to time series analysis within the preset period to obtain the travel habits of the target vehicle in different time periods, such as going to work at a fixed destination during the morning time period on weekdays; the candidate driving sections of the target vehicle in the preset time period are analyzed according to the travel habits, and the congestion level of the target vehicle when passing through the candidate driving sections is analyzed by accessing the navigation data and pushed, and the optimal driving route is changed based on the congestion level; the driving feedback data of the target vehicle after the push in the preset time period is read to update the travel habits.

[0080] Figure 4 The block diagram of the driving trajectory extraction and processing system based on driving recorder data of the present invention is shown.

[0081] The second aspect of the present invention further provides a driving trajectory extraction and processing system 4 based on driving recorder data, the system comprising: a memory 41, a processor 42, the memory comprising a driving trajectory extraction and processing method program based on driving recorder data, the driving trajectory extraction and processing method program based on driving recorder data when executed by the processor implements the following steps:

[0082] Obtaining road data of the target city, preprocessing the road data to generate a city road network structure diagram, reading GPS data of the target vehicle according to the driving recorder data, matching the GPS trajectory sequence with the city road network structure diagram and converting it into a road network-level trajectory sequence;

[0083] Determine the driving scene of the target vehicle according to the trajectory sequence at the road network level, identify other vehicles in the driving scene based on the driving recorder data of the target vehicle, and perform vehicle-to-vehicle relationship analysis through the track data of other vehicles;

[0084] Constructing a risk assessment model for driving scenarios, evaluating the risk distribution of driving scenarios based on the vehicle-to-vehicle relationship and driving comfort according to the risk assessment model, predicting the risk distribution at a preset time in the future, and obtaining a changing trend of the risk distribution;

[0085] The early warning information is generated based on the risk change trend, and the sampling frequency and enhanced data of the driving recorder data are adaptively increased to update the risk distribution of the driving scene in real time.

[0086] The third aspect of the present invention also provides a computer-readable storage medium, which includes a driving trajectory extraction and processing method program based on driving recorder data. When the driving trajectory extraction and processing method program based on driving recorder data is executed by a processor, the steps of the driving trajectory extraction and processing method based on driving recorder data as described in any one of the above items are implemented.

[0087] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0088] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0089] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0090] A person of ordinary skill in the art can understand that: all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: a mobile storage device, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0091] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0092] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for extracting and processing driving trajectories based on driving recorder data, characterized in that: The following steps are involved: Obtaining road data of the target city, preprocessing the road data to generate a city road network structure diagram, reading GPS data of the target vehicle according to the driving recorder data, matching the GPS trajectory sequence with the city road network structure diagram and converting it into a road network-level trajectory sequence; Determine the driving scene of the target vehicle according to the trajectory sequence at the road network level, identify other vehicles in the driving scene based on the driving recorder data of the target vehicle, and perform vehicle-to-vehicle relationship analysis through the track data of other vehicles; Constructing a risk assessment model for driving scenarios, evaluating the risk distribution of driving scenarios based on the vehicle-to-vehicle relationship and driving comfort according to the risk assessment model, predicting the risk distribution at a preset time in the future, and obtaining a changing trend of the risk distribution; Generate early warning information based on the risk change trend, and adaptively increase the sampling frequency and enhanced data of the driving recorder data to update the risk distribution of the driving scene in real time; The GPS data of the target vehicle is read according to the data from the driving recorder, and the GPS trajectory sequence is matched with the urban road network structure diagram and converted into a road network-level trajectory sequence, specifically: Obtaining the road network data of the target city, simplifying the road network data, retaining the road network data of the preset road grade, gridding the road network data, and guiding the merging of intersections of the same grid according to the connectivity relationship of the intersections in the road network; The rationality of the merged road network is verified based on traffic expertise and real urban road traffic conditions. Intersections are extracted as nodes from the simplified road network data, and the road sections between intersections are used as edge structures. The edge attributes of the edge structures are set to construct the urban road network structure diagram. Extracting the dashcam data of the vehicle in the historical time period, separating the GPS track points in the dashcam data, generating a GPS track sequence in combination with the time sequence, segmenting the GPS track sequence, and removing the track segments without movement; Mapping the segmented trajectory segments to the urban road network structure diagram, setting a matching window to calculate the similarity between the turning point and the intersection position, and when the position similarity between the adjacent turning points and the corresponding intersection meets the preset standard, determining whether the shortest path between the intersections is greater than a preset distance threshold; When the distance is greater than the preset distance threshold, matching points are selected in the path between the intersections to set the matching window for re-matching, traversing all the intersections in the urban road network structure diagram, and converting the GPS data into a trajectory sequence at the road network level; Carry out workshop relationship analysis through the track data of other vehicles, specifically: In the driving scene, a reference coordinate system is constructed according to the underlying reference information, vertex coordinates and centroid coordinates of anchor frames of other vehicles are obtained in the reference coordinate system, and driving trajectory fitting of other vehicles is performed based on the vertex coordinates and centroid coordinates; The centroid coordinates are used as tracking corner points, the centroid coordinates are diluted at preset time intervals, the driving trajectory is obtained and matched with the corresponding vehicle, and the driving trajectory in the current driving scene is visualized; Acquire motion features according to the driving trajectories of the target vehicle and other vehicles, encode the relative positions of the target vehicle and other vehicles in the driving scene at different times, combine the relative position codes with the motion features, and construct a motion feature vector; Adding a spatial position label to the motion feature vector based on the reference coordinate system, calculating a spatial attention weight for the motion feature vector with the label, extracting the attention degree of the target vehicle in space, and weighting the motion feature vector using the spatial attention weight; The weighted motion feature vector is used to obtain interaction information of other vehicles through convolution and maximum pooling operations, and the vehicle-to-vehicle relationship with the target vehicle is determined based on the interaction information; A risk assessment model for driving scenarios is constructed, and the risk distribution of driving scenarios is assessed based on the vehicle-to-vehicle relationship and driving comfort according to the risk assessment model, specifically: Acquire motion features from a trajectory sequence of a road network level in a preset historical period of the target vehicle, cluster the trajectory sequence using the motion features, and cluster different sequence segments to generate a number of sequence segment subsets; Acquire the driving recorder data of each sequence segment in the sequence segment subset, determine the corresponding driving scene, perform redundancy removal processing on the driving scene, match the motion feature with the processed driving scene, and use the motion feature as the driving preference of each driving scene; Retrieving the corresponding driving preference according to the current driving scene, and calculating the similarity between the motion characteristics of the current driving scene and the driving preference to obtain a driving comfort evaluation; Analyze risk factors according to accident instances in driving scenarios corresponding to historical preset time periods, match the risk factors with vehicle-to-vehicle relationships, and construct a risk assessment model for driving scenarios; Obtain the Euclidean distance between the target vehicle and other vehicles in the driving scene, calculate the weight information of other vehicles based on the Euclidean distance and the driving comfort of the target vehicle, and extract parameters based on the risk factors through the driving trajectories of other vehicles; In the risk assessment model, risk analysis is performed based on risk parameters and weight information, and the risks of other vehicles are spatially differentiated in the driving scene to visualize the risk distribution.

2. The method for extracting and processing driving trajectories based on driving recorder data according to claim 1, characterized in that: The driving scene of the target vehicle is determined according to the trajectory sequence at the road network level, and other vehicles in the driving scene are identified based on the driving recorder data of the target vehicle, specifically: Using the road network-level trajectory sequence to perform driving positioning, and determining the traffic facility resources at the positioning location in the target city using the driving recorder data and traffic monitoring data stored in the cloud according to the driving positioning; Constructing underlying reference information of a driving scene based on the traffic facility resources, reading real-time driving recorder data of a target vehicle to generate a video frame sequence of the driving scene, and annotating the video frame sequence using the underlying reference information; Other vehicles are identified in the video frame sequence. The Yolov3 algorithm is used as the backbone extraction network. The preprocessed video frame images are imported into the backbone extraction network to construct a feature pyramid to obtain a multi-scale feature map. Introducing a spatial attention mechanism and a channel attention mechanism, assigning weight information to the multi-scale feature map to update the feature distribution, recalibrating the feature map, and fusing the recalibrated feature maps by aggregating context feature maps of different ranges to obtain an optimized feature expression; The optimized feature expression and the preset texture features and shape features are fused to generate multi-dimensional fusion features in the video scene, and the anchor frame is calibrated according to the multi-dimensional fusion features to output the target information and position information of other vehicles.

3. The method for extracting and processing driving trajectories based on driving recorder data according to claim 1, characterized in that: The early warning information is generated by the risk change trend, and the sampling frequency and enhanced data of the driving recorder data are adaptively increased, specifically: Select the risk distribution of the previous n moments in the current driving scene to construct a risk distribution vector for a preset time slice, import the risk vector into the Bi-LSTM network, and use hidden state update to obtain the risk distribution change trend; Obtaining the risk distribution after a preset time according to the change trend, and generating warning information if the risk corresponding to other vehicles is greater than a preset threshold; The deviation information between the risk degree and the preset threshold is obtained to evaluate the warning level of the warning information, and the warning level is weighted based on the meteorological data. The weighted warning level is used to adaptively change the sampling frequency of the driving recorder data and the enhanced data of the driving recorder screen.

4. A driving trajectory extraction and processing system based on driving recorder data, characterized in that: The system includes: a memory and a processor, wherein the memory includes a driving trajectory extraction and processing method program based on driving recorder data, and when the driving trajectory extraction and processing method program based on driving recorder data is executed by the processor, the following steps are implemented: Obtaining road data of the target city, preprocessing the road data to generate a city road network structure diagram, reading GPS data of the target vehicle according to the driving recorder data, matching the GPS trajectory sequence with the city road network structure diagram and converting it into a road network-level trajectory sequence; Determine the driving scene of the target vehicle according to the trajectory sequence at the road network level, identify other vehicles in the driving scene based on the driving recorder data of the target vehicle, and perform vehicle-to-vehicle relationship analysis through the track data of other vehicles; Constructing a risk assessment model for driving scenarios, evaluating the risk distribution of driving scenarios based on the vehicle-to-vehicle relationship and driving comfort according to the risk assessment model, predicting the risk distribution at a preset time in the future, and obtaining a changing trend of the risk distribution; Generate early warning information based on the risk change trend, and adaptively increase the sampling frequency and enhanced data of the driving recorder data to update the risk distribution of the driving scene in real time; The GPS data of the target vehicle is read according to the data from the driving recorder, and the GPS trajectory sequence is matched with the urban road network structure diagram and converted into a road network-level trajectory sequence, specifically: Obtaining the road network data of the target city, simplifying the road network data, retaining the road network data of the preset road grade, gridding the road network data, and guiding the merging of intersections of the same grid according to the connectivity relationship of the intersections in the road network; The rationality of the merged road network is verified based on traffic expertise and real urban road traffic conditions. Intersections are extracted as nodes from the simplified road network data, and the road sections between intersections are used as edge structures. The edge attributes of the edge structures are set to construct the urban road network structure diagram. Extracting the dashcam data of the vehicle in the historical time period, separating the GPS track points in the dashcam data, generating a GPS track sequence in combination with the time sequence, segmenting the GPS track sequence, and removing the track segments without movement; Mapping the segmented trajectory segments to the urban road network structure diagram, setting a matching window to calculate the similarity between the turning point and the intersection position, and when the position similarity between the adjacent turning points and the corresponding intersection meets the preset standard, determining whether the shortest path between the intersections is greater than a preset distance threshold; When the distance is greater than the preset distance threshold, matching points are selected in the path between the intersections to set the matching window for re-matching, traversing all the intersections in the urban road network structure diagram, and converting the GPS data into a trajectory sequence at the road network level; Carry out workshop relationship analysis through the track data of other vehicles, specifically: In the driving scene, a reference coordinate system is constructed according to the underlying reference information, vertex coordinates and centroid coordinates of anchor frames of other vehicles are obtained in the reference coordinate system, and driving trajectory fitting of other vehicles is performed based on the vertex coordinates and centroid coordinates; The centroid coordinates are used as tracking corner points, the centroid coordinates are diluted at preset time intervals, the driving trajectory is obtained and matched with the corresponding vehicle, and the driving trajectory in the current driving scene is visualized; Acquire motion features according to the driving trajectories of the target vehicle and other vehicles, encode the relative positions of the target vehicle and other vehicles in the driving scene at different times, combine the relative position codes with the motion features, and construct a motion feature vector; Adding a spatial position label to the motion feature vector based on the reference coordinate system, calculating a spatial attention weight for the motion feature vector with the label, extracting the attention degree of the target vehicle in space, and weighting the motion feature vector using the spatial attention weight; The weighted motion feature vector is used to obtain interaction information of other vehicles through convolution and maximum pooling operations, and the vehicle-to-vehicle relationship with the target vehicle is determined based on the interaction information; A risk assessment model for driving scenarios is constructed, and the risk distribution of driving scenarios is assessed based on the vehicle-to-vehicle relationship and driving comfort according to the risk assessment model, specifically: Acquire motion features from a trajectory sequence of a road network level in a preset historical period of the target vehicle, cluster the trajectory sequence using the motion features, and cluster different sequence segments to generate a number of sequence segment subsets; Acquire the driving recorder data of each sequence segment in the sequence segment subset, determine the corresponding driving scene, perform redundancy removal processing on the driving scene, match the motion feature with the processed driving scene, and use the motion feature as the driving preference of each driving scene; Retrieving the corresponding driving preference according to the current driving scene, and calculating the similarity between the motion characteristics of the current driving scene and the driving preference to obtain a driving comfort evaluation; Analyze risk factors according to accident instances in driving scenarios corresponding to historical preset time periods, match the risk factors with vehicle-to-vehicle relationships, and construct a risk assessment model for driving scenarios; Obtain the Euclidean distance between the target vehicle and other vehicles in the driving scene, calculate the weight information of other vehicles based on the Euclidean distance and the driving comfort of the target vehicle, and extract parameters based on the risk factors through the driving trajectories of other vehicles; In the risk assessment model, risk analysis is performed based on risk parameters and weight information, and the risks of other vehicles are spatially differentiated in the driving scene to visualize the risk distribution.

5. The driving trajectory extraction and processing system based on driving recorder data according to claim 4 is characterized in that: The driving scene of the target vehicle is determined according to the trajectory sequence at the road network level, and other vehicles in the driving scene are identified based on the driving recorder data of the target vehicle, specifically: Using the road network-level trajectory sequence to perform driving positioning, and determining the traffic facility resources at the positioning location in the target city using the driving recorder data and traffic monitoring data stored in the cloud according to the driving positioning; Constructing underlying reference information of a driving scene based on the traffic facility resources, reading real-time driving recorder data of a target vehicle to generate a video frame sequence of the driving scene, and annotating the video frame sequence using the underlying reference information; Other vehicles are identified in the video frame sequence. The Yolov3 algorithm is used as the backbone extraction network. The preprocessed video frame images are imported into the backbone extraction network to construct a feature pyramid to obtain a multi-scale feature map. Introducing a spatial attention mechanism and a channel attention mechanism, assigning weight information to the multi-scale feature map to update the feature distribution, recalibrating the feature map, and fusing the recalibrated feature maps by aggregating context feature maps of different ranges to obtain an optimized feature expression; The optimized feature expression and the preset texture features and shape features are fused to generate multi-dimensional fusion features in the video scene, and the anchor frame is calibrated according to the multi-dimensional fusion features to output the target information and position information of other vehicles.

Images