Training method and device of congestion vehicle identification model and computer program product

Abstract

The present disclosure provides a training method and device of a blocked vehicle identification model, electronic equipment, storage medium and program product, relating to the technical field of artificial intelligence, and especially suitable for the technical field of automatic driving. The specific implementation scheme is: obtaining a training sample set, the training sample including features of state nodes in a running track of an obstacle, features of road nodes corresponding to the running track, and a state label representing whether the obstacle is a blocked vehicle; performing feature coding on the state nodes and the road nodes to obtain respective coding features of the state nodes and the road nodes; performing information transmission between the obstacle and the road based on the respective coding features of the two kinds of nodes to obtain post-transmission features of the state nodes; and training a blocked vehicle identification model by using a machine learning method, taking the post-transmission features as input and taking the state label corresponding to the post-transmission features as expected output. The present disclosure improves the accuracy of the model and the applicability of the model to various blocked vehicle identification scenarios.

Description

Technical Field

[0001] The present disclosure relates to the fields of artificial intelligence technology, specifically to the fields of autonomous driving and deep learning technology, and more particularly to a training method and device for a blocked vehicle recognition model, as well as a blocked vehicle recognition method, device, electronic device, storage medium, and computer program product, which can be used in autonomous driving scenarios. Background Art

[0002] The task of blocking vehicle recognition involves automatically identifying stationary vehicles that are parked for a long time or in queues, thereby enabling autonomous vehicles to predict the status of surrounding vehicles. In reality, blocking vehicle recognition scenarios are complex and diverse, requiring numerous judgment signals. Methods based on manually designed rules struggle to adapt to these diverse scenarios and suffer from low accuracy. Summary of the Invention

[0003] The present disclosure provides a training method and device for a blocked vehicle identification model, as well as a blocked vehicle identification method and device, an electronic device, a storage medium, and a computer program product.

[0004] According to a first aspect, a method for training a blocking vehicle recognition model is provided, comprising: obtaining a training sample set, wherein the training samples in the training sample set include features of state nodes in the obstacle's trajectory, features of road nodes of the road corresponding to the trajectory, and a state label indicating whether the obstacle is a blocking vehicle; encoding the features of the state nodes and road nodes using an encoding module to obtain encoding features corresponding to each of the state nodes and the road nodes; transferring information between the obstacle and the road using a local attention aggregation module based on the encoding features corresponding to each of the state nodes and the road nodes to obtain post-transfer features corresponding to the state nodes; and training the blocking vehicle recognition model using a machine learning method with the post-transfer features as input and the state labels corresponding to the post-transfer features as desired output.

[0005] According to a second aspect, a method for identifying a blocked vehicle is provided, comprising: determining features of a state node in an obstacle's trajectory and features of a road node on a road corresponding to the trajectory; encoding the features of the state node and the road node using an encoding module to obtain encoding features corresponding to each of the state node and the road node; transferring information between the obstacle and the road based on the encoding features corresponding to each of the state node and the road node using a local attention aggregation module to obtain transferred features corresponding to the state node; and determining whether the obstacle is a blocked vehicle based on the transferred features using a trained blocked vehicle identification model.

[0006] According to a third aspect, a training device for a blocking vehicle recognition model is provided, comprising: an acquisition unit configured to acquire a training sample set, wherein the training samples in the training sample set include features of state nodes in the obstacle's running trajectory, features of road nodes of the road corresponding to the running trajectory, and a state label representing whether the obstacle is a blocking vehicle; a first encoding unit configured to perform feature encoding on the state nodes and road nodes via an encoding module to obtain encoding features corresponding to each of the state nodes and the road nodes; a first transfer unit configured to perform information transfer between the obstacle and the road via a local attention aggregation module based on the encoding features corresponding to each of the state nodes and the road nodes to obtain transferred features corresponding to the state nodes; and a training unit configured to train the blocking vehicle recognition model using a machine learning method with the transferred features as input and the state labels corresponding to the transferred features as desired output.

[0007] According to a fourth aspect, a method for identifying a blocked vehicle is provided, comprising: a first determining unit configured to determine features of a state node in a running trajectory of an obstacle and features of a road node of a road corresponding to the running trajectory; a second encoding unit configured to perform feature encoding on the state node and the road node via an encoding module to obtain encoded features corresponding to each of the state node and the road node; a second transmitting unit configured to transmit information between the obstacle and the road via a local attention aggregation module based on the encoded features corresponding to each of the state node and the road node to obtain a transferred feature corresponding to the state node; and the second determining unit configured to determine whether the obstacle is a blocked vehicle based on the transferred feature via a trained blocked vehicle identification model.

[0008] According to the fifth aspect, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so as to enable the at least one processor to execute the method described in any implementation of the first aspect or the second aspect.

[0009] According to a sixth aspect, a non-transitory computer-readable storage medium storing computer instructions is provided, where the computer instructions are used to cause a computer to execute the method described in any one of the implementations of the first and second aspects.

[0010] According to a seventh aspect, a computer program product is provided, comprising: a computer program, which implements the method described in any implementation manner of the first aspect or the second aspect when executed by a processor.

[0011] According to the technology disclosed herein, a method for training a blocked vehicle identification model is provided. During the training process, information is transferred between obstacles and roads based on the coded features corresponding to state nodes and road nodes. The resulting transferred features of the state nodes include both the features of the obstacle itself and the features of the road on which the obstacle is located. This improves the richness and expressiveness of the feature data underlying the blocked vehicle identification model, thereby enhancing the model's accuracy and applicability to various blocked vehicle identification scenarios. The trained blocked vehicle identification model is used to determine whether obstacles surrounding the vehicle are blocked vehicles, thereby improving the accuracy of blocked vehicle identification.

[0012] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings are provided to facilitate a better understanding of the present invention and do not constitute a limitation of the present disclosure.

[0014] Figure 1 is an exemplary system architecture diagram in which an embodiment of the present disclosure may be applied;

[0015] Figure 2 is a flowchart of an embodiment of a method for training a blocked vehicle recognition model according to the present disclosure;

[0016] Figure 3 is a schematic diagram of an obstacle subgraph according to this embodiment;

[0017] Figure 4 is a schematic diagram of a road subgraph according to this embodiment;

[0018] Figure 5 is a schematic diagram of information flow of the blocked vehicle identification model according to this embodiment;

[0019] Figure 6 is a schematic structural diagram of a blocked vehicle identification model according to this embodiment;

[0020] Figure 7 is a schematic diagram of an application scenario of the training method for a blocked vehicle recognition model according to this embodiment;

[0021] Figure 8 is a flowchart of an embodiment of a feature fusion operation of the local attention aggregation module according to this embodiment;

[0022] Figure 9 is a schematic structural diagram of a local attention aggregation module according to this embodiment;

[0023] Figure 10 is a flowchart of another embodiment of a method for training a blocked vehicle recognition model according to the present disclosure;

[0024] Figure 11 is a flow chart of an embodiment of a blocked vehicle identification method according to the present disclosure;

[0025] Figure 12 is a flow chart of another embodiment of a blocked vehicle identification method according to the present disclosure;

[0026] Figure 13 is a structural diagram of an embodiment of a training device for a blocked vehicle identification model according to the present disclosure;

[0027] Figure 14 is a structural diagram of an embodiment of a blocked vehicle identification device according to the present disclosure;

[0028] Figure 15 It is a schematic diagram of the structure of a computer system suitable for implementing the embodiments of the present disclosure. DETAILED DESCRIPTION

[0029] The following description of exemplary embodiments of the present disclosure is made in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding. These details should be considered as merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0030] In the technical solutions disclosed herein, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0031] Figure 1 An exemplary architecture 100 is shown to which the blocked vehicle identification model training method and apparatus, and the blocked vehicle identification method and apparatus of the present disclosure can be applied.

[0032] like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. The communication connections between terminal devices 101, 102, and 103 constitute a topological network, and network 104 is used to provide a medium for communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables.

[0033] The terminal devices 101, 102, and 103 can be hardware devices or software that support network connection to interact with the server and process data. When the terminal devices 101, 102, and 103 are hardware, they can be various electronic devices that support network connection, information acquisition, interaction, display, processing and other functions, including but not limited to data acquisition devices such as sensors, vehicle-mounted computers, smart phones, tablet computers, e-book readers, laptop computers, and desktop computers, etc. When the terminal devices 101, 102, and 103 are software, they can be installed in the electronic devices listed above. They can be implemented as multiple software or software modules for providing distributed services, for example, or they can be implemented as a single software or software module. No specific limitation is made here.

[0034] Server 105 can be a server that provides various services. For example, it can be a backend processing server that uses machine learning to train a blocked vehicle identification model using training sample sets provided by terminal devices 101, 102, and 103. For another example, it can be a backend processing server that uses the trained blocked vehicle model to identify whether an obstacle around terminal devices 101, 102, and 103 is a blocked vehicle. As an example, server 105 can be a cloud server.

[0035] It should be noted that the server can be either hardware or software. When the server is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When the server is software, it can be implemented as multiple software or software modules (e.g., software or software modules for providing distributed services), or as a single software or software module. No specific limitations are given here.

[0036] It should also be noted that the methods for training a blocked vehicle identification model and identifying a blocked vehicle provided in the embodiments of the present disclosure can be executed by a server, a terminal device, or a server and a terminal device in coordination with each other. Accordingly, the various components (e.g., various units) included in the training device and identifying a blocked vehicle identification model can be entirely located in the server, entirely located in the terminal device, or separately located in the server and the terminal device.

[0037] It should be understood that Figure 1 The number of terminal devices, networks, and servers described is merely illustrative. Any number of terminal devices, networks, and servers may be provided as needed. When the method for training a blocked vehicle identification model and the electronic device on which the method is executed do not need to transmit data with other electronic devices, the system architecture may only include the electronic device (e.g., a server or terminal device) on which the method for training a blocked vehicle identification model and the method are executed.

[0038] Please refer to Figure 2 , Figure 2 This is a flow chart of a method for training a blocked vehicle recognition model provided by an embodiment of the present disclosure, wherein process 200 includes the following steps:

[0039] Step 201: Obtain a training sample set.

[0040] In this embodiment, the execution subject of the training method of the blocking vehicle recognition model (for example, Figure 1 The terminal device or server in the training sample set can obtain the training data set remotely or locally via a wired network connection or a wireless network connection. The training samples in the training sample set include the features of the state nodes in the obstacle's running trajectory, the features of the road nodes of the road corresponding to the running trajectory, and the state label indicating whether the obstacle is a blocking vehicle.

[0041] By detecting the movement of obstacles, we can obtain their trajectory. The obstacle's trajectory can be characterized by the state nodes corresponding to the obstacle at different time points. State node features characterize the obstacle in the state represented by the state node, such as its location, size, speed, orientation, and type. Road node features characterize the road point corresponding to the road node, such as road type, number of lanes, and different intersections.

[0042] Blocked vehicles generally include vehicles parked for a long time or vehicles in a queue. In this embodiment, the execution entity can use different state labels to identify whether the obstacle corresponding to the training sample is a blocked vehicle. For example, a state label of "1" indicates that the obstacle is a blocked vehicle, while a state label of "0" indicates that the obstacle is not a blocked vehicle. In this embodiment, the execution entity can further distinguish whether the blocked vehicle is parked for a long time or a queued vehicle using state labels. For example, a state label of "1" indicates that the obstacle is parked for a long time, a state label of "2" indicates that the obstacle is a queued vehicle, and a state label of "0" indicates that the obstacle is not a blocked vehicle.

[0043] In some optional implementations of this embodiment, the characteristics of both state nodes and road nodes include geometric characteristics and attribute characteristics. The geometric characteristics of state nodes include: the obstacle's center point position coordinates, orientation, dimensions (length, width, height), and speed direction in three-dimensional space; the attribute characteristics of state nodes include: speed magnitude, headlight signal, and obstacle type. The geometric characteristics of road nodes include: the position coordinates and orientation of the road center point; the attribute characteristics of road nodes include: the distance from the road center point to the left and right edges of the road, the road type, and the type of the left and right edges of the road.

[0044] Traffic lights are also tied to corresponding road nodes, and whether a road node is in a blind spot (an area that cannot be observed by a vehicle's sensors) is also a binary attribute feature. At this point, the road node's attribute features also include: traffic light color and whether it is in a blind spot.

[0045] Step 202: feature encoding is performed on the state node and the road node by the encoding module to obtain encoding features corresponding to the state node and the road node respectively.

[0046] In this embodiment, the execution entity may perform feature encoding on the state node and the road node through the encoding module to obtain encoding features corresponding to the state node and the road node.

[0047] As an example, for the characteristics of each state node, the above-mentioned execution entity can encode the characteristics of the state node through an encoder to obtain the encoded characteristics of the state node; for the characteristics of each road node, the above-mentioned execution entity can encode the characteristics of the road node through an encoder to obtain the encoded characteristics of the road node.

[0048] In some optional implementations of this embodiment, the encoding module includes a temporal encoding module and a context encoding module. In this implementation, the execution subject may perform step 202 as follows:

[0049] First, for each state node, the state node and the historical state nodes of the state node are temporally encoded by the temporal encoding module to obtain a temporal encoding feature corresponding to each state node and integrating the features of the historical state nodes.

[0050] For each state node, its corresponding historical state node is the state node corresponding to the historical time point before the time point corresponding to the state node. As an example, for each state node, the characteristics of the state node and the characteristics of the historical state nodes within a preset time period from the state node can be time-series encoded to obtain the time-series encoding characteristics corresponding to the state node.

[0051] Second, for each road node, the context coding module performs context coding on the road node and its adjacent road nodes to obtain context coding features corresponding to each road node that integrate features of adjacent road nodes.

[0052] As an example, for each road node, the above-mentioned execution entity can determine the neighboring road nodes within a preset distance range of the road node; then, context-encode the features of the road node and the features of the neighboring road nodes to obtain the context-encoded features of the road node.

[0053] In this implementation, by performing temporal coding on the state nodes, the temporal coding features corresponding to each state node are integrated with the features of the historical state nodes; by performing context coding on the road nodes, the context coding features of each road node are integrated with the features of the adjacent road nodes, thereby further improving the richness and expressiveness of the information in the coding features of the state nodes and road nodes.

[0054] In some optional implementations of this embodiment, the execution entity may perform the first step as follows: first, for each state node, an obstacle subgraph corresponding to the state node, including directed state nodes, is generated based on the state node and a preset number of historical state nodes corresponding to the state node; then, a temporal coding module is used to temporally encode features of the state nodes in the obstacle subgraph to obtain temporal coding features corresponding to each state node.

[0055] Continue to refer Figure 3 , shows a schematic diagram of an obstacle subgraph. Obstacle subgraph 300 includes state nodes 301-305. State nodes 301-304 are historical state nodes of state node 305. The direction of the directed state nodes in the obstacle subgraph is "301->302->303->304->305," representing the trajectory of the obstacle.

[0056] The temporal coding module may adopt a self-attention module, and the preset number may be flexibly set according to the actual situation (for example, the computing power of the execution subject). For example, the preset number is 10.

[0057] In this implementation, the obstacle subgraph is different from the traditional image grid network. Information processing is performed through the obstacle subgraph, a vector network. While ensuring the richness of the coding features, it reduces the amount of calculation and improves the processing efficiency of the algorithm.

[0058] In some optional implementations of this embodiment, the above-mentioned execution entity can perform the above-mentioned second step in the following manner: first, for each road node, based on the road node and a preset number of adjacent road nodes corresponding to the road node, generate a road subgraph corresponding to the road node, including directed road nodes; and use a context coding module to context-code the coding features corresponding to the road nodes in the road subgraph to obtain context-coded features corresponding to each road node.

[0059] Continue to refer Figure 4 , which shows a schematic diagram of a road subgraph. The road subgraph 400 includes road nodes 401-411. The context encoding module may employ a self-attention module.

[0060] In this implementation, the road subgraph is different from the traditional image raster network. Information processing is performed through the vector network of the road subgraph. While ensuring the richness of the coding features, the amount of calculation is reduced and the processing efficiency of the algorithm is improved.

[0061] In step 203, the local attention aggregation module is used to transfer information between the obstacle and the road based on the encoding features corresponding to the state node and the road node, and obtain the transferred features corresponding to the state node.

[0062] In this embodiment, the above-mentioned execution entity can use the LAA (Local Attentive Aggregation) module to transfer information between obstacles and roads based on the encoding features corresponding to the state nodes and road nodes, and obtain the transferred features corresponding to the state nodes.

[0063] As an example, for each state node of an obstacle, the execution entity can determine at least one road node that is spatially adjacent to the state node. The execution entity then uses the local attention fusion module to fuse the encoding features of the state node with the encoding features of at least one adjacent road node to obtain the transferred features of the state node. In this example, the execution entity can also use the local attention fusion module to fuse the temporal encoding features of the state node with the contextual encoding features of at least one adjacent road node to obtain the transferred features of the state node.

[0064] As another example, the execution entity can perform multiple information transfers between the obstacle and the road. Specifically, first, for each state node of the obstacle, the execution entity can determine at least one road node that is spatially adjacent to the state node; for each road node of the road, the execution entity can determine at least one state node that is spatially adjacent to the road node. Then, for each road node, the local attention fusion module fuses the coding features of the road node with the coding features of at least one state node adjacent to the road node to obtain the first post-transfer features of the road node, thereby achieving the first information transfer from the road to the obstacle; finally, for each state node, the local attention fusion module fuses the coding features of the state node with the post-transfer features of at least one road node adjacent to the state node to obtain the post-transfer features of the state node, thereby achieving the first information transfer from the obstacle to the road. In this example, the coding features of the state node can also be temporal coding features, and the coding features of the road node can also be contextual coding features.

[0065] Continue to refer Figure 5, which shows an information flow diagram of a blocked vehicle identification model. The information flow 500 includes a temporal encoding process 501, information transfer processes 502-503, a context encoding process 504, and an information transfer process 505.

[0066] Continue to refer Figure 6 , showing the Figure 5 The corresponding block vehicle identification model structure diagram. The processing process 601-605 of each structure in the block vehicle identification model 600 is the same as Figure 5 The processes 501-505 correspond to each other.

[0067] In some optional implementations of this embodiment, the execution entity may perform step 203 as follows:

[0068] First, for each road node, the local attention aggregation module is used to fuse the features of the road node and the temporal coding features of the state nodes in the neighborhood of the road node to obtain the first fused coding features of the road node.

[0069] In this implementation, for each road node, the execution entity can determine a specified number of state nodes in the road node's neighborhood. Then, through the local attention aggregation module, the features of the road node are fused with the temporal coding features of the specified number of state nodes in the road node's neighborhood to obtain the first fused coding feature of the road node, thereby completing the first information transfer process from the obstacle to the road. The specified number can be set based on actual conditions, for example, 16.

[0070] Second, for each road node, based on the local attention aggregation module, the first fused coded features of the road node and the first fused coded features of the road nodes in the neighborhood of the road node are fused to obtain the second fused coded features of the road node.

[0071] In this implementation, for each road node, the execution entity can determine a specified number of road nodes in the neighborhood of the road node; then, through the local attention aggregation module, the first fused coded features of the road node and the first fused coded features of the road nodes in the neighborhood of the road node are fused to obtain the second fused coded features of the road node, thereby completing the second road-to-road information transmission process.

[0072] In this implementation, the execution entity executes step 202 as follows: first, for each road node, a road subgraph corresponding to the road node and including directed road nodes is generated based on the road node and a preset number of adjacent road nodes corresponding to the road node; then, the context coding module performs context coding on the second fused coding features corresponding to the road nodes in the road subgraph to obtain context coding features corresponding to each road node.

[0073] In this implementation, context coding of the road subgraph is performed based on the second fused coding features of the road nodes obtained from the first information transfer and the second information transfer to obtain context coding features of each road node.

[0074] In this implementation, the above step 203 also includes: for each state node, based on the local attention aggregation module, fusing the temporal coding features of the state node and the contextual coding features of the road nodes in the neighborhood of the state node to obtain the post-transfer features corresponding to each state node.

[0075] For each state node, the execution entity can determine a specified number of road nodes in the neighborhood of the state node. Then, based on the local attention aggregation module, the temporal coding features of the state node and the contextual coding features of the specified number of road nodes in the neighborhood of the state node are fused to obtain the post-transmission features corresponding to each state node, thereby completing the third information transmission process from the road to the obstacle.

[0076] In this implementation, a specific implementation method for the three-step information transmission process from obstacle to road, road to road, and road to obstacle is provided, which further improves the richness and expressiveness of the post-transmission features of the final state node.

[0077] For the information transmission process based on the local attention fusion module, the above-mentioned execution entity can fuse the corresponding features based on addition, weighted fusion, etc.

[0078] In step 204 , a machine learning method is used to train a blocked vehicle recognition model using the transferred features as input and the state labels corresponding to the transferred features as expected output.

[0079] In this embodiment, the execution subject may utilize a machine learning method, with the transferred features as input and the state labels corresponding to the transferred features as expected output, to train a blocked vehicle recognition model.

[0080] As an example, the aforementioned execution entity uses the transferred features as input to the initial blocked vehicle identification model, obtaining an actual output to determine whether an obstacle is a blocked vehicle. The loss between the actual output and the expected output is then determined based on a preset loss function, and the gradient is determined based on the loss. The initial blocked vehicle identification model is then updated using stochastic gradient descent. The aforementioned training operation is iterated, and upon reaching a preset termination condition, a trained blocked vehicle identification model is obtained. Thus, the encoding module, the local attention aggregation module, and the blocked vehicle identification model are combined to form a complete blocked vehicle identification system. This blocked vehicle identification system is also known as a DNGNN (Directed Node Graph Neural Network) model.

[0081] The recognition module can employ an MLP (Multilayer Perceptron) to decode and predict the actual output based on the post-transmission characteristics of the state nodes. Preset termination conditions can include, for example, training time exceeding a preset time threshold, training times exceeding a preset number threshold, or training loss convergence.

[0082] Continue to see Figure 7 , Figure 7 FIG. 700 is a schematic diagram of an application scenario of the training method of the blocked vehicle recognition model according to this embodiment. Figure 7 In the application scenario, the server used to train the blocked vehicle recognition model first obtains a training sample set. The training sample set 701 includes features 7011 of the state node in the obstacle's trajectory, features 7012 of the road node corresponding to the trajectory, and a state label 7013 indicating whether the obstacle is a blocked vehicle. The encoding module then performs feature encoding on the state node and road node, obtaining the encoded features 702 corresponding to each state node and road node. The local attention aggregation module then performs information transfer between the obstacle and the road based on the encoded features corresponding to each state node and road node. Specifically, two information transfer processes are performed: from obstacle to road and from road to obstacle, ultimately obtaining the post-transfer features 703 corresponding to the state node. Using machine learning methods, the post-transfer features are used as input and the state labels corresponding to the post-transfer features are used as the desired output to train the blocked vehicle recognition model.

[0083] This embodiment provides a training method for a blocked vehicle recognition model. During the training process, information is transferred between obstacles and roads based on the coded features corresponding to state nodes and road nodes. The resulting transferred features of the state nodes include both the features of the obstacle itself and the features of the road on which the obstacle is located. This improves the richness and expressiveness of the feature data underlying the blocked vehicle recognition model, thereby enhancing the model's accuracy and applicability to various blocked vehicle recognition scenarios. The trained blocked vehicle recognition model determines whether obstacles surrounding the vehicle are blocked vehicles, thereby improving the accuracy of blocked vehicle recognition.

[0084] Continue to refer Figure 8 , shows a schematic process 800 of an embodiment of a feature fusion operation of a local attention aggregation module according to this embodiment, including the following steps:

[0085] Step 801: For a target node to be subjected to feature fusion, determine a specified number of neighboring nodes corresponding to the target node.

[0086] In this embodiment, the execution subject of the feature fusion operation (for example, Figure 1 The server or terminal device in the embodiment can determine a specified number of neighboring nodes corresponding to the target node for which feature fusion is to be performed.

[0087] For the first feature fusion operation of the first information transmission process from the obstacle to the road, the second feature fusion operation of the second information transmission process from the road to the road, and the third feature fusion operation of the third information transmission process from the road to the obstacle, the target nodes are the road node, the road node, and the status node, and their adjacent nodes are the status node, the road node, and the road node.

[0088] The preset number can be set according to actual conditions. For example, the preset number is 16.

[0089] Step 802: For each neighboring node, determine the fusion weight corresponding to the neighboring node according to the geometric features of the neighboring node relative to the target node through the relative pose encoder.

[0090] In this embodiment, the execution entity may determine, for each neighboring node, a fusion weight corresponding to the neighboring node according to the geometric features of the neighboring node relative to the target node through a relative pose encoder.

[0091] Neither the geometric features of the state nodes nor the geometric features of the road nodes participate in the fusion of features in information transmission, but are only used to determine the fusion weights.

[0092] As an example, based on the spatial position and orientation of the target node, the relative position Rt and relative orientation Ra of the neighboring nodes relative to the target node are calculated. Furthermore, the relative position Rt and relative orientation Ra are encoded using the RPE (Relative Pose Encoder) module to obtain the corresponding fusion weights.

[0093] Step 803 : Based on the updated attribute features and fusion weights corresponding to each neighboring node and the pre-fusion features currently corresponding to the target node, a fused feature corresponding to the target node is obtained.

[0094] In this embodiment, the execution entity may obtain the fused features corresponding to the target node based on the updated attribute features and fusion weights corresponding to each neighboring node and the pre-fusion features currently corresponding to the target node.

[0095] Among them, the updated attribute features in each feature fusion operation are the temporal coding features of the state node, the first fused coding features of the road node and the context coding features of the road node, respectively; the features before fusion are the attribute features of the road node, the first fused coding features of the road node and the temporal coding features of the state node in each feature fusion operation, and the features after fusion are the first fused coding features of the road node, the second fused coding features of the road node and the corresponding transferred features of the state node in each feature fusion operation.

[0096] As an example, for each neighboring node, the above-mentioned execution entity multiplies the updated attribute features corresponding to the neighboring node and the fusion weight to obtain the weighted attribute features corresponding to the neighboring node; then, the weighted attribute features of each neighboring node and the pre-fusion features currently corresponding to the target node are fused to obtain the fusion features corresponding to the target node.

[0097] In this embodiment, a specific fusion method of a local attention aggregation module is provided. Since the fusion weight is determined based on the geometric features of the neighboring nodes relative to the target node, no matter how the vehicle turns, the local attention aggregation module can conveniently perform feature fusion based on the relative geometric features. That is, the local attention aggregation module has rotation invariance, and the local attention aggregation module has multi-target parallel processing capabilities, which improves the feature fusion efficiency and accuracy during information transmission.

[0098] Continue to refer Figure 9 , showing the structural diagram of the local attention aggregation module.

[0099] In some optional implementations of this embodiment, the execution entity may perform step 803 in the following manner: First, for each neighboring node, perform the following operations:

[0100] First, the weighted attribute features corresponding to the neighboring node are obtained according to the updated attribute features and fusion weights corresponding to the neighboring node; then, the output features of the neighboring node are obtained by combining the weighted attribute features and the updated attribute features corresponding to the neighboring node.

[0101] Second, the output features of each neighboring node and the pre-fusion features corresponding to the target node are combined to obtain the post-fusion features corresponding to the target node.

[0102] As an example, the execution entity may use MLP to perform feature fusion.

[0103] Specifically, in the first feature fusion operation of information transmission from the obstacle to the road, for each state node (neighboring node) adjacent to the road node (target node), the fusion weight is determined according to the geometric features of the state node relative to the road node; according to the time series coding features (updated attribute features) and fusion weights corresponding to each state node, the weighted attribute features corresponding to each state node are obtained; for each state node, the weighted attribute features and time series coding features of the state node are combined to obtain the output features corresponding to the state node; the output features of each state node and the attribute features of the road node (features before fusion) are combined to obtain the first fused coding features (fused features) corresponding to the road node.

[0104] In the second feature fusion operation of information transmission from road to road, for each road node (neighboring node) adjacent to a road node (target node), a fusion weight is determined based on the geometric features of the neighboring road node relative to the road node as the target node; based on the first fused coded features (updated attribute features) and fusion weight corresponding to each neighboring road node, the weighted attribute features corresponding to each neighboring road node are obtained; for each neighboring road node, the weighted attribute features and the first fused coded features of the neighboring road node are combined to obtain the output features corresponding to the state node; and the second fused coded features (fused features) corresponding to the road node are obtained by combining the output features of each state node and the first fused coded features (pre-fusion features) of the road node.

[0105] In the third feature fusion operation of information transmission from the road to the obstacle, for each road node (neighboring node) adjacent to the state node (target node), the fusion weight is determined based on the geometric features of the road node relative to the state node; based on the context coding features (updated attribute features) corresponding to each road node and the fusion weight, the weighted attribute features corresponding to each road node are obtained; for each adjacent road node, the weighted attribute features and context coding features of the road node are combined to obtain the output features corresponding to the road node; the output features of each road node and the temporal coding features of the state node (features before fusion) are combined to obtain the transferred features (fused features) corresponding to the state node.

[0106] In this implementation, the weighted attribute features and updated attribute features corresponding to the neighboring nodes are further fused, thereby further improving the feature fusion degree and richness of the final transferred features.

[0107] Continue to refer Figure 10 , shows a schematic process 1000 of another embodiment of a training method for a blocked vehicle recognition model according to the present disclosure, comprising the following steps:

[0108] Step 1001: Obtain a training sample set.

[0109] The training samples in the training sample set include features of state nodes in the running trajectory of the obstacle, features of road nodes of the road corresponding to the running trajectory, and a state label representing whether the obstacle is a blocking vehicle.

[0110] Step 1002 : For each state node, generate an obstacle subgraph corresponding to the state node and including directed state nodes according to the state node and a preset number of historical state nodes corresponding to the state node.

[0111] Step 1003 : Time-series coding is performed on the features of the state nodes in the obstacle subgraph by using a time-series coding module to obtain a time-series coding feature corresponding to each state node.

[0112] Step 1004: For each road node, the local attention aggregation module is used to fuse the features of the road node and the temporal coding features of the state nodes in the neighborhood of the road node to obtain a first fused coding feature of the road node.

[0113] Step 1005: For each road node, based on the local attention aggregation module, the first fused coded feature of the road node and the first fused coded features of the road nodes in the neighborhood of the road node are fused to obtain the second fused coded feature of the road node.

[0114] Step 1006 : For each road node, generate a road subgraph corresponding to the road node and including directed road nodes based on the road node and a preset number of adjacent road nodes corresponding to the road node.

[0115] Step 1007 : context-encoding the second fused coded features corresponding to the road nodes in the road subgraph using a context-encoding module to obtain context-encoded features corresponding to each road node.

[0116] Step 1008: For each state node, based on the local attention aggregation module, the temporal coding features of the state node and the contextual coding features of the road nodes in the neighborhood of the state node are fused to obtain the post-transfer features corresponding to each state node.

[0117] In step 1009 , a machine learning method is used to train a blocked vehicle recognition model using the transferred features as input and the state labels corresponding to the transferred features as expected output.

[0118] It can be seen from this embodiment that Figure 2 Compared to the corresponding embodiments, process 1000 of the method for training a blocked vehicle recognition model in this embodiment specifically illustrates the feature encoding process and the information transmission process between obstacles and roads, further improving the richness and expressiveness of the data on which the blocked vehicle recognition model is based, thereby improving the accuracy of the model and the applicability of the model to various blocked vehicle recognition scenarios.

[0119] Continue to refer Figure 11 , shows a schematic process 1100 of an embodiment of a blocked vehicle identification method according to the present disclosure, comprising the following steps:

[0120] Step 1101 : Determine the characteristics of the state nodes in the running trajectory of the obstacle and the characteristics of the road nodes of the road corresponding to the running trajectory.

[0121] In this embodiment, the execution subject of the blocking vehicle identification method (for example, Figure 1 The terminal device or server in the system can determine the characteristics of the state node in the running trajectory of the obstacle and the characteristics of the road node of the road corresponding to the running trajectory.

[0122] For example, for a moving vehicle, the execution entity may determine the characteristics of state nodes in the trajectory of obstacles surrounding the vehicle and the characteristics of road nodes on the road corresponding to the trajectory. Specifically, the execution entity may determine the characteristics of state nodes in the obstacle's trajectory and the characteristics of road nodes on the road corresponding to the trajectory based on information collection devices such as radar, cameras, and sensors.

[0123] In some optional implementations of this embodiment, the execution entity may perform step 1101 as follows:

[0124] First, based on a preset time interval, a target data frame representing each state node is determined from the data representing the running trajectory of the obstacle, and the characteristics of the state node are determined based on the target data frame.

[0125] In this implementation, the preset time interval can be set based on actual conditions. For example, the preset time interval is 0.2 seconds. Each target data frame represents a state node, and features of the corresponding state node are obtained by extracting features from the target data frame. The data includes, but is not limited to, video data collected by a camera and sensor data collected by various sensors (e.g., a speed sensor).

[0126] Second, based on the preset sampling interval, each road node and its characteristics are determined from the high-precision map corresponding to the running trajectory.

[0127] In this implementation, the preset sampling interval can be set based on actual conditions. For example, the preset sampling interval is 5 meters. A road node is determined from the high-precision map at every preset sampling interval, and the characteristics of the road node are determined based on the high-precision map.

[0128] In this implementation, a specific implementation method for determining the characteristics of the state node and the characteristics of the road node is provided, which improves the accuracy of the characteristics while ensuring the accuracy of the determined characteristics.

[0129] Step 1102: feature encoding is performed on the state node and the road node by the encoding module to obtain encoding features corresponding to the state node and the road node respectively.

[0130] In some optional implementations of this embodiment, the encoding module includes a temporal encoding module and a context encoding module. In this implementation, the execution entity may perform step 1102 as follows: First, for each state node, the temporal encoding module performs temporal encoding on the state node and its historical state nodes to obtain a temporal encoding feature corresponding to each state node that incorporates features of the historical state nodes; Second, for each road node, the context encoding module performs context encoding on the road node and its neighboring road nodes to obtain a context encoding feature corresponding to each road node that incorporates features of the neighboring road nodes.

[0131] In some optional implementations of this embodiment, the execution entity may perform the first step as follows: first, for each state node, an obstacle subgraph corresponding to the state node, including directed state nodes, is generated based on the state node and a preset number of historical state nodes corresponding to the state node; then, a temporal coding module is used to temporally encode features of the state nodes in the obstacle subgraph to obtain temporal coding features corresponding to each state node.

[0132] In step 1103, the local attention aggregation module is used to transfer information between the obstacle and the road based on the encoding features corresponding to the state node and the road node, and the transferred features corresponding to the state node are obtained.

[0133] In some optional implementations of this embodiment, the execution entity may perform step 1103 as follows:

[0134] First, for each road node, the local attention aggregation module is used to fuse the features of the road node and the temporal coding features of the state nodes in the neighborhood of the road node to obtain the first fused coding features of the road node.

[0135] Second, for each road node, based on the local attention aggregation module, the first fused coded features of the road node and the first fused coded features of the road nodes in the neighborhood of the road node are fused to obtain the second fused coded features of the road node.

[0136] In this implementation, the execution entity may perform step 1102 as follows: first, for each road node, a road subgraph including directed road nodes is generated corresponding to the road node based on the road node and a preset number of adjacent road nodes corresponding to the road node; then, the context coding module performs context coding on the second fused coding features corresponding to the road nodes in the road subgraph to obtain context coding features corresponding to each road node.

[0137] In this implementation, the above step 1103 also includes: for each state node, based on the local attention aggregation module, fusing the temporal coding features of the state node and the contextual coding features of the road nodes in the neighborhood of the state node to obtain the post-transfer features corresponding to each state node.

[0138] In some optional implementations of this embodiment, the local attention aggregation module performs feature fusion operations in the following manner: first, for the target node to be feature fused, determine a specified number of neighboring nodes corresponding to the target node; second, for each neighboring node, determine the fusion weight corresponding to the neighboring node according to the geometric features of the neighboring node relative to the target node through the relative pose encoder; third, based on the updated attribute features and fusion weights corresponding to each neighboring node and the pre-fusion features currently corresponding to the target node, obtain the fused features corresponding to the target node, wherein the updated attribute features in each feature fusion operation are the temporal coding features of the state node, the first fused coding features of the road node, and the contextual coding features of the road node, respectively; the pre-fusion features in each feature fusion operation are the attribute features of the road node, the first fused coding features of the road node, and the temporal coding features of the state node, respectively; the fused features in each feature fusion operation are the first fused coding features of the road node, the second fused coding features of the road node, and the transferred features corresponding to the state node, respectively.

[0139] In some optional implementations of this embodiment, the above-mentioned execution entity can perform the above-mentioned third step in the following manner: first, for each neighboring node, perform the following operations: according to the updated attribute features and fusion weights corresponding to the neighboring node, obtain the weighted attribute features corresponding to the neighboring node; combine the weighted attribute features and updated attribute features corresponding to the neighboring node to obtain the output features of the neighboring node; second, combine the output features of each neighboring node and the pre-fusion features currently corresponding to the target node to obtain the fusion features corresponding to the target node.

[0140] Step 1104 : Determine whether the obstacle is a blocking vehicle based on the transferred features using the trained blocking vehicle recognition model.

[0141] In this embodiment, the upper execution entity can also more specifically determine whether the blocked vehicle is a long-parked vehicle or a queued vehicle, and assist the downstream PnC (Planning and Control) in making corresponding driving decisions based on the identification results, thereby achieving more intelligent driving of the main vehicle.

[0142] In this embodiment, the above-mentioned execution entity can implement this embodiment with reference to embodiments 200 and 800 and the corresponding implementation methods therein, which will not be repeated here.

[0143] In this embodiment, a blocked vehicle identification method is provided, which improves the accuracy and efficiency of blocked vehicle identification results based on a blocked vehicle identification model.

[0144] Continue to refer Figure 12, shows a schematic process 1200 of another embodiment of a blocked vehicle identification method according to the present disclosure, comprising the following steps:

[0145] Step 1201 : Determine the characteristics of the state nodes in the running trajectory of the obstacle and the characteristics of the road nodes of the road corresponding to the running trajectory.

[0146] Step 1202 : For each state node, generate an obstacle subgraph corresponding to the state node and including directed state nodes according to the state node and a preset number of historical state nodes corresponding to the state node.

[0147] Step 1203 : Time-series coding is performed on the features of the state nodes in the obstacle subgraph by using a time-series coding module to obtain a time-series coding feature corresponding to each state node.

[0148] Step 1204: For each road node, the local attention aggregation module is used to fuse the features of the road node and the temporal coding features of the state nodes in the neighborhood of the road node to obtain a first fused coding feature of the road node.

[0149] Step 1205: For each road node, based on the local attention aggregation module, the first fused coded feature of the road node and the first fused coded features of the road nodes in the neighborhood of the road node are fused to obtain the second fused coded feature of the road node.

[0150] Step 1206 : For each road node, generate a road subgraph corresponding to the road node and including directed road nodes based on the road node and a preset number of adjacent road nodes corresponding to the road node.

[0151] Step 1207 : context-encoding the second fused coded features corresponding to the road nodes in the road subgraph using a context-encoding module to obtain context-encoded features corresponding to each road node.

[0152] In step 1208, for each state node, based on the local attention aggregation module, the temporal coding features of the state node and the contextual coding features of the road nodes in the neighborhood of the state node are fused to obtain the post-transfer features corresponding to each state node.

[0153] Step 1209 : Determine whether the obstacle is a blocking vehicle based on the transferred features using the trained blocking vehicle recognition model.

[0154] It can be seen from this embodiment that Figure 11Compared with the corresponding embodiment, the process 1200 of the blocked vehicle identification method in this embodiment specifically illustrates the feature encoding process and the information transmission process between the obstacle and the road, further improving the richness and expressiveness of the data based on the blocked vehicle identification model, thereby further improving the accuracy of the model's recognition results.

[0155] Continue to refer Figure 13 As an implementation of the methods shown in the above figures, the present disclosure provides an embodiment of a training device for a blocked vehicle recognition model. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0156] like Figure 13 As shown, the training apparatus 1300 for a blocking vehicle recognition model includes: an acquisition unit 1301, configured to acquire a training sample set, wherein the training samples in the training sample set include features of state nodes in the obstacle's running trajectory, features of road nodes of the road corresponding to the running trajectory, and a state label indicating whether the obstacle is a blocking vehicle; a first encoding unit 1302, configured to perform feature encoding on the state nodes and road nodes through an encoding module to obtain encoding features corresponding to each of the state nodes and the road nodes; a first transfer unit 1303, configured to perform information transfer between the obstacle and the road based on the encoding features corresponding to each of the state nodes and the road nodes through a local attention aggregation module to obtain a transferred feature corresponding to the state node; and a training unit 1304, configured to utilize a machine learning method, with the transferred feature as input and the state label corresponding to the transferred feature as the desired output, to train a blocking vehicle recognition model.

[0157] In some optional implementations of this embodiment, the encoding module includes a temporal encoding module and a context encoding module, and a first encoding unit 1302, and is further configured to: for each state node, perform temporal encoding on the state node and the historical state nodes of the state node through the temporal encoding module to obtain a temporal encoding feature corresponding to each state node that integrates the features of the historical state nodes; for each road node, perform context encoding on the road node and the adjacent road nodes of the road node through the context encoding module to obtain a context encoding feature corresponding to each road node that integrates the features of the adjacent road nodes.

[0158] In some optional implementations of this embodiment, the first encoding unit 1302 is further configured to: for each state node, generate an obstacle subgraph corresponding to the state node, including directed state nodes, based on the state node and a preset number of historical state nodes corresponding to the state node; and perform time-series encoding on the features of the state nodes in the obstacle subgraph through the time-series encoding module to obtain time-series encoding features corresponding to each state node.

[0159] In some optional implementations of this embodiment, the first transfer unit 1303 is further configured to: for each road node, fuse the features of the road node and the temporal coding features of the state nodes in the neighborhood of the road node through the local attention aggregation module to obtain a first fused coding feature of the road node; for each road node, fuse the first fused coding feature of the road node and the first fused coding features of the road nodes in the neighborhood of the road node based on the local attention aggregation module to obtain a second fused coding feature of the road node; and the first encoding unit 1302 is further configured to: for each road node, generate a road subgraph corresponding to the road node, including directed road nodes, based on the road node and a preset number of neighboring road nodes corresponding to the road node; perform context coding on the second fused coding features corresponding to the road nodes in the road subgraph through the context coding module to obtain a context coding feature corresponding to each road node; and the first transfer unit 1303 is further configured to: for each state node, fuse the temporal coding features of the state node and the context coding features of the road nodes in the neighborhood of the state node based on the local attention aggregation module to obtain a transferred feature corresponding to each state node.

[0160] In some optional implementations of this embodiment, the features of the state node and the features of the road node both include geometric features and attribute features; and the first transfer unit 1303 is further configured to: perform a feature fusion operation through the local attention aggregation module in the following manner: for a target node to be feature fused, determine a specified number of neighboring nodes corresponding to the target node; for each neighboring node, determine a fusion weight corresponding to the neighboring node based on the geometric features of the neighboring node relative to the target node through a relative pose encoder; based on the updated attribute features and fusion weight corresponding to each neighboring node and the pre-fusion features currently corresponding to the target node, obtain a fused feature corresponding to the target node, wherein the updated attribute features in each feature fusion operation are sequentially the temporal coding features of the state node, the first fused coding features of the road node, and the contextual coding features of the road node; the pre-fusion features in each feature fusion operation are sequentially the attribute features of the road node, the first fused coding features of the road node, and the temporal coding features of the state node; and the fused features in each feature fusion operation are sequentially the first fused coding features of the road node, the second fused coding features of the road node, and the transferred features corresponding to the state node.

[0161] In some optional implementations of this embodiment, the first transmission unit 1303 is further configured to: for each neighboring node, perform the following operations: obtain the weighted attribute features corresponding to the neighboring node based on the updated attribute features and fusion weights corresponding to the neighboring node; obtain the output features of the neighboring node by combining the weighted attribute features and updated attribute features corresponding to the neighboring node; and obtain the fused features corresponding to the target node by combining the output features of each neighboring node and the pre-fusion features currently corresponding to the target node.

[0162] This embodiment provides a training device for a blocked vehicle recognition model. During the training process, information is transferred between obstacles and roads based on the coded features corresponding to state nodes and road nodes. The resulting transferred features of the state nodes include both the features of the obstacle itself and the features of the road on which the obstacle is located. This improves the richness and expressiveness of the feature data underlying the blocked vehicle recognition model, thereby enhancing the model's accuracy and applicability to various blocked vehicle recognition scenarios. The trained blocked vehicle recognition model determines whether obstacles surrounding the vehicle are blocked vehicles, thereby improving the accuracy of blocked vehicle recognition.

[0163] Continue to refer Figure 14 As an implementation of the methods shown in the above figures, the present disclosure provides an embodiment of a method for identifying a blocked vehicle. Figure 11 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0164] like Figure 14 As shown, the blocking vehicle identification method 1400 includes: a first determination unit 1401, configured to determine the characteristics of the state node in the running trajectory of the obstacle and the characteristics of the road node of the road corresponding to the running trajectory; a second encoding unit 1402, configured to encode the characteristics of the state node and the road node through the encoding module to obtain the encoding characteristics corresponding to each of the state node and the road node; a second transfer unit 1403, configured to transfer information between the obstacle and the road based on the encoding characteristics corresponding to each of the state node and the road node through the local attention aggregation module to obtain the transferred characteristics corresponding to the state node; and a second determination unit 1404, configured to determine whether the obstacle is a blocking vehicle based on the transferred characteristics using the trained blocking vehicle identification model.

[0165] In some optional implementations of this embodiment, the encoding module includes a temporal encoding module and a context encoding module, and a second encoding unit 1402, and is further configured to: for each state node, perform temporal encoding on the state node and the historical state nodes of the state node through the temporal encoding module to obtain a temporal encoding feature corresponding to each state node that integrates the features of the historical state nodes; for each road node, perform context encoding on the road node and the adjacent road nodes of the road node through the context encoding module to obtain a context encoding feature corresponding to each road node that integrates the features of the adjacent road nodes.

[0166] In some optional implementations of this embodiment, the second encoding unit 1402 is further configured to: for each state node, generate an obstacle subgraph corresponding to the state node, including directed state nodes, based on the state node and a preset number of historical state nodes corresponding to the state node; and perform time-series encoding on the features of the state nodes in the obstacle subgraph through the time-series encoding module to obtain time-series encoding features corresponding to each state node.

[0167] In some optional implementations of this embodiment, the second transfer unit 1403 is further configured to: for each road node, fuse the features of the road node and the temporal coding features of the state nodes in the neighborhood of the road node through the local attention aggregation module to obtain a first fused coding feature of the road node; for each road node, fuse the first fused coding feature of the road node and the first fused coding features of the road nodes in the neighborhood of the road node based on the local attention aggregation module to obtain a second fused coding feature of the road node; and the second encoding unit 1402 is further configured to: for each road node, generate a road subgraph corresponding to the road node, including directed road nodes, based on the road node and a preset number of neighboring road nodes corresponding to the road node; perform context coding on the second fused coding features corresponding to the road nodes in the road subgraph through the context coding module to obtain a context coding feature corresponding to each road node; and the second transfer unit 1403 is further configured to: for each state node, fuse the temporal coding features of the state node and the context coding features of the road nodes in the neighborhood of the state node based on the local attention aggregation module to obtain a transferred feature corresponding to each state node.

[0168] In some optional implementations of this embodiment, the features of the state node and the features of the road node both include geometric features and attribute features; and the second transfer unit 1403 is further configured to: perform a feature fusion operation through the local attention aggregation module in the following manner: for a target node to be feature fused, determine a specified number of neighboring nodes corresponding to the target node; for each neighboring node, determine a fusion weight corresponding to the neighboring node based on the geometric features of the neighboring node relative to the target node through a relative pose encoder; based on the updated attribute features and fusion weight corresponding to each neighboring node and the pre-fusion features currently corresponding to the target node, obtain a fused feature corresponding to the target node, wherein the updated attribute features in each feature fusion operation are sequentially the temporal coding features of the state node, the first fused coding features of the road node, and the contextual coding features of the road node; the pre-fusion features in each feature fusion operation are sequentially the attribute features of the road node, the first fused coding features of the road node, and the temporal coding features of the state node; and the fused features in each feature fusion operation are sequentially the first fused coding features of the road node, the second fused coding features of the road node, and the transferred features corresponding to the state node.

[0169] In some optional implementations of this embodiment, the second transmission unit 1403 is further configured to: for each neighboring node, perform the following operations: obtain the weighted attribute features corresponding to the neighboring node based on the updated attribute features and fusion weights corresponding to the neighboring node; combine the weighted attribute features and updated attribute features corresponding to the neighboring node to obtain the output features of the neighboring node; and combine the output features of each neighboring node and the pre-fusion features currently corresponding to the target node to obtain the fusion features corresponding to the target node.

[0170] In some optional implementations of this embodiment, the first determination unit 1403 is further configured to: determine the target data frame representing each state node from the data representing the running trajectory of the obstacle based on a preset time interval, and determine the characteristics of the state node based on the target data frame; determine each road node and the characteristics of each road node from the high-precision map corresponding to the running trajectory based on a preset sampling interval.

[0171] In this embodiment, a blocked vehicle identification device is provided, which improves the accuracy and efficiency of blocked vehicle identification results based on a blocked vehicle identification model.

[0172] According to an embodiment of the present disclosure, the present disclosure further provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can implement the training method for a blocked vehicle recognition model and the blocked vehicle recognition method described in any of the above embodiments when executing the instructions.

[0173] According to an embodiment of the present disclosure, the present disclosure further provides a readable storage medium storing computer instructions, which are used to enable a computer to implement the training method of the blocked vehicle recognition model and the blocked vehicle recognition method described in any of the above embodiments when executed.

[0174] The embodiments of the present disclosure provide a computer program product. When executed by a processor, the computer program can implement the training method of the blocked vehicle recognition model and the blocked vehicle recognition method described in any of the above embodiments.

[0175] Figure 15A schematic block diagram of an example electronic device 1500 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are provided as examples only and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0176] like Figure 15 As shown, device 1500 includes a computing unit 1501, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1502 or a computer program loaded from a storage unit 1508 into a random access memory (RAM) 1503. Various programs and data required for the operation of device 1500 can also be stored in RAM 1503. Computing unit 1501, ROM 1502, and RAM 1503 are connected to each other via a bus 1504. An input / output (I / O) interface 1505 is also connected to bus 1504.

[0177] Various components in device 1500 are connected to I / O interface 1505, including an input unit 1506, such as a keyboard and mouse; an output unit 1507, such as various types of displays and speakers; a storage unit 1508, such as a magnetic disk and optical disk; and a communication unit 1509, such as a network card, a modem, a wireless communication transceiver, etc. Communication unit 1509 allows device 1500 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0178] Computing unit 1501 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of computing unit 1501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Computing unit 1501 performs the various methods and processes described above, such as the method for training a blocked vehicle identification model. For example, in some embodiments, the method for training a blocked vehicle identification model can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as storage unit 1508. In some embodiments, part or all of the computer program can be loaded and / or installed onto device 1500 via ROM 1502 and / or communication unit 1509. When the computer program is loaded into RAM 1503 and executed by computing unit 1501, one or more steps of the method for training a blocked vehicle identification model described above can be performed. Alternatively, in other embodiments, the computing unit 1501 may be configured to execute the training method of the blocked vehicle recognition model in any other appropriate manner (for example, by means of firmware).

[0179] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0180] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0181] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0182] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0183] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0184] A computer system may include a client and a server. The client and server are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, establishing a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host, a host product within a cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosts and virtual private servers (VPS). It may also be a server in a distributed system or a server integrated with blockchain.

[0185] According to the technical solution of the embodiments of the present disclosure, a method for training a blocked vehicle identification model is provided. During the training process, information is transferred between obstacles and roads based on the coded features corresponding to state nodes and road nodes. The resulting transferred features of the state nodes include both the features of the obstacle itself and the features of the road on which the obstacle is located. This improves the richness and expressiveness of the feature data underlying the blocked vehicle identification model, thereby enhancing the model's accuracy and applicability to various blocked vehicle identification scenarios. The trained blocked vehicle identification model is used to determine whether obstacles surrounding the vehicle are blocked vehicles, thereby improving the accuracy of blocked vehicle identification.

[0186] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions provided by this disclosure can be achieved. This is not a limitation herein.

[0187] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the scope of protection of this disclosure.

Claims

1. A training method for a blocked vehicle recognition model, comprising: Acquire a training sample set, wherein the training samples in the training sample set include features of state nodes in the running trajectory of the obstacle, features of road nodes of the road corresponding to the running trajectory, and a state label indicating whether the obstacle is a blocking vehicle; Performing feature encoding on the state nodes and the road nodes by a coding module to obtain coding features corresponding to the state nodes and the road nodes, respectively, including: for each state node, performing time-series encoding on the state node and its historical state nodes by a time-series encoding module in the coding module to obtain a time-series encoding feature corresponding to each state node that integrates features of the historical state nodes; for each road node, performing context encoding on the road node and its neighboring road nodes by a context encoding module in the coding module to obtain a context-series encoding feature corresponding to each road node that integrates features of neighboring road nodes; The method includes: for each state node, fusing the temporal coding features of the state node and the context coding features of the road nodes in the neighborhood of the state node, based on the local attention aggregation module, to obtain the post-transfer features corresponding to the state node; A machine learning method is used to train a blocked vehicle recognition model with the transferred features as input and the state labels corresponding to the transferred features as expected output.

2. The method according to claim 1, wherein For each state node, the state node and the historical state nodes of the state node are time-series encoded by the time-series encoding module to obtain a time-series encoding feature corresponding to each state node and integrated with the features of the historical state nodes, including: For each state node, generating an obstacle subgraph corresponding to the state node and including directed state nodes based on the state node and a preset number of historical state nodes corresponding to the state node; The time sequence coding module performs time sequence coding on the features of the state nodes in the obstacle subgraph to obtain the time sequence coding features corresponding to each state node.

3. The method according to claim 1, wherein The local attention aggregation module transfers information between the obstacle and the road based on the coded features corresponding to the state node and the road node, thereby obtaining a transferred feature corresponding to the state node, further comprising: For each road node, the local attention aggregation module is used to fuse the features of the road node and the temporal coding features of the state nodes in the neighborhood of the road node to obtain a first fused coding feature of the road node; For each road node, based on the local attention aggregation module, the first fused coded feature of the road node and the first fused coded features of the road nodes in the neighborhood of the road node are fused to obtain a second fused coded feature of the road node; and For each road node, context coding is performed on the road node and its neighboring road nodes by the context coding module to obtain a context coding feature corresponding to each road node and integrating features of the neighboring road nodes, including: For each road node, generating a road subgraph corresponding to the road node and including directed road nodes based on the road node and a preset number of adjacent road nodes corresponding to the road node; The context coding module performs context coding on the second fused coding features corresponding to the road nodes in the road subgraph to obtain context coding features corresponding to each road node.

4. The method according to claim 3, wherein: The characteristics of the state node and the characteristics of the road node both include geometric characteristics and attribute characteristics; and The local attention aggregation module performs feature fusion operations in the following way: For a target node to be feature fused, determining a specified number of neighboring nodes corresponding to the target node; For each neighboring node, determine the fusion weight corresponding to the neighboring node according to the geometric features of the neighboring node relative to the target node through the relative pose encoder; Based on the updated attribute features and fusion weights corresponding to each neighboring node and the pre-fusion features currently corresponding to the target node, the fused features corresponding to the target node are obtained, wherein the updated attribute features are, in each feature fusion operation, the temporal coding features of the state node, the first fused coding features of the road node, and the context coding features of the road node; the pre-fusion features are, in each feature fusion operation, the attribute features of the road node, the first fused coding features of the road node, and the temporal coding features of the state node; the fused features are, in each feature fusion operation, the first fused coding features of the road node, the second fused coding features of the road node, and the transferred features corresponding to the state node.

5. The method according to claim 4, wherein The method of obtaining the fused features corresponding to the target node based on the updated attribute features and fusion weights corresponding to each neighboring node and the pre-fusion features currently corresponding to the target node includes: For each neighboring node, do the following: Obtain the weighted attribute features corresponding to the neighboring node according to the updated attribute features and fusion weights corresponding to the neighboring node; Combining the weighted attribute feature and the updated attribute feature corresponding to the neighboring node to obtain an output feature of the neighboring node; and The output features of each neighboring node and the pre-fusion features currently corresponding to the target node are combined to obtain the post-fusion features corresponding to the target node.

6. A method for identifying a blocked vehicle, comprising: Determining characteristics of a state node in a running trajectory of the obstacle and characteristics of a road node of a road corresponding to the running trajectory; Performing feature encoding on the state nodes and the road nodes by a coding module to obtain coding features corresponding to the state nodes and the road nodes, respectively, including: for each state node, performing time-series encoding on the state node and its historical state nodes by a time-series encoding module in the coding module to obtain a time-series encoding feature corresponding to each state node that integrates features of the historical state nodes; for each road node, performing context encoding on the road node and its neighboring road nodes by a context encoding module in the coding module to obtain a context-series encoding feature corresponding to each road node that integrates features of neighboring road nodes; The method includes: for each state node, fusing the temporal coding features of the state node and the context coding features of the road nodes in the neighborhood of the state node, based on the local attention aggregation module, to obtain the post-transfer features corresponding to the state node; The trained blocking vehicle recognition model is used to determine whether the obstacle is a blocking vehicle based on the transferred features.

7. The method according to claim 6, wherein: For each state node, the state node and the historical state nodes of the state node are time-series encoded by the time-series encoding module to obtain a time-series encoding feature corresponding to each state node and integrated with the features of the historical state nodes, including: For each state node, generating an obstacle subgraph corresponding to the state node and including directed state nodes based on the state node and a preset number of historical state nodes corresponding to the state node; The time sequence coding module performs time sequence coding on the features of the state nodes in the obstacle subgraph to obtain the time sequence coding features corresponding to each state node.

8. The method according to claim 6, wherein: The local attention aggregation module transfers information between the obstacle and the road based on the coded features corresponding to the state node and the road node, thereby obtaining a transferred feature corresponding to the state node, further comprising: For each road node, the local attention aggregation module is used to fuse the features of the road node and the temporal coding features of the state nodes in the neighborhood of the road node to obtain a first fused coding feature of the road node; For each road node, based on the local attention aggregation module, the first fused coded feature of the road node and the first fused coded features of the road nodes in the neighborhood of the road node are fused to obtain a second fused coded feature of the road node; and For each road node, context coding is performed on the road node and its neighboring road nodes by the context coding module to obtain a context coding feature corresponding to each road node and integrating features of the neighboring road nodes, including: For each road node, generating a road subgraph corresponding to the road node and including directed road nodes based on the road node and a preset number of adjacent road nodes corresponding to the road node; The context coding module performs context coding on the second fused coding features corresponding to the road nodes in the road subgraph to obtain context coding features corresponding to each road node.

9. The method according to claim 8, wherein The characteristics of the state node and the characteristics of the road node both include geometric characteristics and attribute characteristics; and The local attention aggregation module performs feature fusion operations in the following way: For a target node to be feature fused, determining a specified number of neighboring nodes corresponding to the target node; For each neighboring node, determine the fusion weight corresponding to the neighboring node according to the geometric features of the neighboring node relative to the target node through the relative pose encoder; Based on the updated attribute features and fusion weights corresponding to each neighboring node and the pre-fusion features currently corresponding to the target node, the fused features corresponding to the target node are obtained, wherein the updated attribute features are, in each feature fusion operation, the temporal coding features of the state node, the first fused coding features of the road node, and the context coding features of the road node; the pre-fusion features are, in each feature fusion operation, the attribute features of the road node, the first fused coding features of the road node, and the temporal coding features of the state node; the fused features are, in each feature fusion operation, the first fused coding features of the road node, the second fused coding features of the road node, and the transferred features corresponding to the state node.

10. The method according to claim 9, wherein: The method of obtaining the fused features corresponding to the target node based on the updated attribute features and fusion weights corresponding to each neighboring node and the pre-fusion features currently corresponding to the target node includes: For each neighboring node, do the following: Obtain the weighted attribute features corresponding to the neighboring node according to the updated attribute features and fusion weights corresponding to the neighboring node; Combining the weighted attribute feature and the updated attribute feature corresponding to the neighboring node to obtain an output feature of the neighboring node; and The output features of each neighboring node and the pre-fusion features currently corresponding to the target node are combined to obtain the post-fusion features corresponding to the target node.

11. The method according to claim 6, wherein: The determining of the characteristics of the state nodes in the running trajectory of the obstacle and the characteristics of the road nodes of the road corresponding to the running trajectory includes: Determining target data frames representing each state node from data representing the movement trajectory of the obstacle based on a preset time interval, and determining features of the state node based on the target data frames; Based on a preset sampling interval, each road node and the characteristics of each road node are determined from the high-precision map corresponding to the running trajectory.

12. A training device for a blocked vehicle recognition model, comprising: an acquisition unit configured to acquire a training sample set, wherein the training samples in the training sample set include features of state nodes in the running trajectory of the obstacle, features of road nodes of the road corresponding to the running trajectory, and a state label indicating whether the obstacle is a blocking vehicle; The first encoding unit is configured to perform feature encoding on the state nodes and the road nodes through an encoding module to obtain encoding features corresponding to the state nodes and the road nodes, including: for each state node, performing time-series encoding on the state node and its historical state nodes through a time-series encoding module in the encoding module to obtain a time-series encoding feature corresponding to each state node that is integrated with features of the historical state nodes; for each road node, performing context encoding on the road node and its neighboring road nodes through a context encoding module in the encoding module to obtain a context-series encoding feature corresponding to each road node that is integrated with features of neighboring road nodes; The first transfer unit is configured to transfer information between the obstacle and the road based on the coding features corresponding to the state node and the road node through a local attention aggregation module to obtain a transferred feature corresponding to the state node, including: for each state node, fusing the temporal coding feature of the state node with the contextual coding features of the road nodes in the state node's neighborhood based on the local attention aggregation module to obtain a transferred feature corresponding to each state node; The training unit is configured to use a machine learning method, take the transferred features as input, and take the state labels corresponding to the transferred features as expected output, to train a blocked vehicle recognition model.

13. A blocked vehicle identification device, comprising: a first determining unit configured to determine features of a state node in a running trajectory of the obstacle and features of a road node of a road corresponding to the running trajectory; The second encoding unit is configured to perform feature encoding on the state nodes and the road nodes through an encoding module to obtain encoding features corresponding to the state nodes and the road nodes, including: for each state node, performing time-series encoding on the state node and its historical state nodes through a time-series encoding module in the encoding module to obtain a time-series encoding feature corresponding to each state node that is integrated with features of the historical state nodes; for each road node, performing context encoding on the road node and its neighboring road nodes through a context encoding module in the encoding module to obtain a context-encoded feature corresponding to each road node that is integrated with features of neighboring road nodes; The second transfer unit is configured to transfer information between the obstacle and the road based on the coding features corresponding to the state node and the road node through a local attention aggregation module to obtain a transferred feature corresponding to the state node, including: for each state node, fusing the temporal coding feature of the state node with the contextual coding features of the road nodes in the state node's neighborhood based on the local attention aggregation module to obtain a transferred feature corresponding to each state node; The second determining unit is configured to determine whether the obstacle is a blocking vehicle based on the transferred features using a trained blocking vehicle recognition model.

14. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 11.

15. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 11.

16. A computer program product comprising: A computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 11.

Images