Lane and path prediction model construction method, path prediction method and device

Abstract

This invention discloses a method, method, and apparatus for constructing a lane and path prediction model, belonging to the field of path prediction. The method includes: acquiring the historical paths of all lanes and vehicles in a target scene; constructing multi-precision lanes based on the lanes; obtaining a set of different lane segments traversed by the vehicle based on the multi-precision lanes, the historical paths, and a fully connected layer, and obtaining multiple predicted paths based on the different lane segment sets; acquiring the vehicle's actual path and the lanes it actually traversed; and obtaining a trained lane and path prediction model based on the multiple predicted paths and the lane segments traversed by the vehicle, along with the actual path and the lanes it actually traversed, using a cross-entropy loss function and a mean squared error loss function.

Description

Technical Field

[0001] This invention belongs to the field of path prediction, and particularly relates to a method for constructing a lane and path prediction model, a path prediction method, and an apparatus. Background Technology

[0002] Path prediction refers to predicting a vehicle's future path based on its historical paths. It has wide applications in robotic navigation, intelligent transportation systems, driver assistance systems, and autonomous driving systems. Lanes, as a crucial component of roads, directly influence vehicle movement by restricting driving space and defining driving directions. Currently, existing path prediction methods typically input lane information as auxiliary features into models such as recurrent neural networks, graph neural networks, or generative adversarial networks to model the interaction between vehicles and lanes. Some methods capture the weight of lane influence on vehicle trajectory through attention mechanisms or directly regress future trajectory point coordinates using an encoder-decoder structure, thereby predicting the vehicle's future path.

[0003] However, existing methods still suffer from the following technical problems: First, existing methods typically employ a fixed-scale lane representation, which cannot simultaneously account for the constraints imposed by lanes on vehicles at different travel distances. When the vehicle's travel distance changes, a single-scale lane representation struggles to depict the multi-level impact of lanes on the vehicle's path, resulting in insufficient modeling ability for lane constraints. Second, most existing methods directly regress the coordinates of the predicted path, lacking explicit utilization of lane structure. This allows the predicted path to deviate from the lane centerline or even cross lane boundaries, leading to prediction results that do not reflect the actual driving behavior of vehicles in the lane. Third, existing methods lack supervision signals for lane segment prediction during model training, making it difficult to guide the model to learn the vehicle's driving decision-making process at the lane level, thus reducing the model's interpretability and prediction accuracy. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method for constructing a lane and path prediction model, comprising: Obtain the historical paths of all lanes and vehicles in the target scene; Based on the described lanes, construct multi-precision lanes; Based on the multi-precision lanes, the historical paths, and the fully connected layer, a set of different lane segments traversed by the vehicle is obtained, and multiple predicted paths are obtained based on the different lane segment sets. Obtain the vehicle's actual path and the lanes it actually passed through; Based on the various predicted paths and the lane segments traversed by the vehicle, and the actual paths and lanes actually traversed, a trained lane and path prediction model is obtained using the cross-entropy loss function and the mean squared error loss function.

[0005] Optionally, based on the lane, a multi-precision lane is constructed, including: Set at least two lane separation lengths; Each lane is segmented according to its respective lane segmentation length to obtain a set of lane segments corresponding to each lane segmentation length and the adjacency relationship between lane segments; The multi-precision lane is obtained by combining the set of lane segments corresponding to each lane segment length and their adjacent relationships.

[0006] Optionally, based on the multi-precision lanes, the historical path, and the fully connected layer, a set of different lane segments traversed by the vehicle is obtained, including: Based on the historical path and the fully connected layer, extract the historical features of the vehicle; Based on the multi-precision lane and the fully connected layer, extract the multi-precision lane features; Based on the historical features, the multi-precision lane features, and the fully connected layer, the probability of the vehicle passing through each lane segment is obtained; Multiple samples are taken based on the given probability to obtain the set of different lane segments.

[0007] Optionally, based on the historical path and the fully connected layer, historical features of the vehicle are extracted, including: The historical path is input into the fully connected layer, and the historical features are output after mapping by the fully connected layer.

[0008] Optionally, multi-precision lane features are extracted based on the multi-precision lanes and the fully connected layer, including: Each lane segment in the multi-precision lane is input into the fully connected layer, and the multi-precision lane features corresponding to each lane segment are output through mapping by the fully connected layer.

[0009] Optionally, based on the historical features, the multi-precision lane features, and the fully connected layer, the probability of the vehicle traversing each lane segment is obtained, including: Calculate the product of the historical features and each corresponding element in the multi-precision lane features to obtain the fused features; The fused features are input into the fully connected layer, mapped by the fully connected layer, and then converted into the probabilities by the softmax function.

[0010] Optionally, multiple samplings are performed based on the probability to obtain the set of different lane segments, including: Based on the probability, multiple sets of lane segments are obtained by sampling, and each set of lane segments contains at least one lane segment.

[0011] Optionally, based on the multiple predicted paths and the lane segments traversed by the vehicle, and the actual paths and lanes actually traversed, a trained lane and path prediction model is obtained using the cross-entropy loss function and the mean squared error loss function, including: The first loss value is calculated using the cross-entropy loss function based on the probability of the vehicle passing through each lane segment and the lane actually passed. Based on the multiple predicted paths and the actual path, the mean square error between each predicted path and the actual path is calculated, and the minimum mean square error is selected as the second loss value. The total loss value is obtained based on the first loss value and the second loss value; The parameters of the fully connected layer are adjusted based on the total loss value to obtain the trained lane and path prediction model.

[0012] To address the aforementioned technical problems, the present invention also provides a path prediction method, comprising: Obtain the historical paths of all lanes and vehicles in the target scene; Set at least two lane segmentation lengths, segment each lane according to each lane segmentation length, obtain the lane segment set corresponding to each lane segmentation length and the adjacency relationship between lane segments, combine the lane segment set corresponding to each lane segmentation length and the adjacency relationship to obtain a multi-precision lane; Historical features of the vehicle are extracted based on the historical path and the fully connected layer. Multi-precision lane features are extracted based on the multi-precision lane and the fully connected layer. The probability of the vehicle passing through each lane segment is obtained based on the historical features, the multi-precision lane features, and the fully connected layer. Multiple samplings are performed based on the stated probability to obtain multiple lane segment sets, and corresponding prediction paths are generated based on each lane segment set, thereby obtaining multiple prediction paths for the vehicle.

[0013] On the other hand, the present invention also provides an electronic device including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0014] On the other hand, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0015] Compared with the prior art, the present invention has the following advantages and technical effects: First, by constructing multi-precision lanes and dividing lanes into equal-length segments at different segmentation scales, the model can simultaneously capture the constraints of lanes on vehicle paths at different spatial granularities, effectively improving the model's ability to model multi-level lane structures. Second, by extracting historical path features and multi-precision lane segment features through fully connected layers, and predicting the probability of a vehicle passing through each lane segment based on the fused features, the predicted path is always generated based on lane segments, ensuring that the prediction results conform to lane geometric constraints and effectively avoiding the problem of predicted paths deviating from lanes or crossing lane boundaries. Third, by jointly using the cross-entropy loss function to supervise lane segment prediction and the mean squared error loss function to supervise path coordinate regression, the model learns both lane-level driving decisions and coordinate-level path generation during training, improving the model's interpretability and prediction accuracy. Attached Figure Description

[0016] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the electronic device structure according to an embodiment of the present invention. Detailed Implementation

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0019] Example 1 like Figure 1 As shown, this embodiment provides a method for constructing a lane and path prediction model, including: Obtain the historical paths of all lanes and vehicles in the target scene; Based on the described lanes, construct multi-precision lanes; Based on the multi-precision lanes, the historical paths, and the fully connected layer, a set of different lane segments traversed by the vehicle is obtained, and multiple predicted paths are obtained based on the different lane segment sets. Obtain the vehicle's actual path and the lanes it actually passed through; Based on the various predicted paths and the lane segments traversed by the vehicle, and the actual paths and lanes actually traversed, a trained lane and path prediction model is obtained using the cross-entropy loss function and the mean squared error loss function.

[0020] The specific process includes: Path prediction refers to predicting a vehicle's future path based on its historical paths. It has wide applications in robotic navigation, intelligent transportation systems, driver assistance systems, and autonomous driving systems. Lanes, as a crucial component of roads, directly influence vehicle movement by limiting driving space and defining driving directions. Therefore, accurately constructing lane impact data and predicting the lanes a vehicle might traverse is of great significance for vehicle trajectory prediction and driving safety.

[0021] In view of this, embodiments of the present invention provide a lane and path prediction model, such as Figure 1 As shown, the model includes: multi-precision lanes, fully connected layers, cross-entropy loss function, and mean squared error loss function. The number of specific network structures contained in each layer of the fully connected layer can be flexibly set according to the construction requirements of the path prediction model. For example, several fully connected layers can be connected sequentially to form the entire fully connected layer of the lane and path prediction model; however, this invention is not limited to this.

[0022] In one embodiment of the present invention, a method for constructing a lane and path prediction model is provided. For example... Figure 2 As shown, the above method specifically includes the following steps: Obtain the historical paths of all lanes and vehicles in the target scene.

[0023] Construct multi-precision lanes using the obtained lane data.

[0024] By utilizing the multi-precision lanes, the vehicle's historical path, and the fully connected layer, a set of different lane segments traversed by the vehicle and multiple predicted paths are obtained.

[0025] The trained lane and path prediction model is obtained by using the cross-entropy loss function and the mean squared error loss function.

[0026] By following the steps above, a well-trained lane and path prediction model can be obtained, which can efficiently represent the future path of the vehicle, making the prediction results consistent with the vehicle's driving behavior in the lane, closer to the real path, and improving the accuracy of the prediction results.

[0027] Optionally, in some embodiments of the present invention, the above-described steps for constructing multi-precision lanes specifically include: Set a lane segment precision (length). Divide each lane into equal-length lane segments according to this precision, and collect all lane segments to form a lane scene with precision, which includes lane adjacency relationships.

[0028] By setting different lane segmentation lengths and repeating the above steps several times, several lane scenarios with varying degrees of precision can be obtained.

[0029] ; in, G This represents a multi-precision lane scenario. U and E These represent the set of lane segments and the adjacency relationship between lane segments, respectively. L This indicates the precision (length) of lane segments.

[0030] Optionally, in some embodiments of the present invention, the above-described steps for constructing multi-precision lanes specifically include: Set a lane segment precision (length). Divide each lane into equal-length lane segments according to this precision, and collect all lane segments to form a lane scene with precision, which includes lane adjacency relationships.

[0031] By setting different lane segmentation lengths and repeating the above steps several times, several lane scenarios with varying degrees of precision can be obtained.

[0032] In this embodiment, the process of constructing multi-precision lanes is specifically represented by the following formula: In the i At each level of precision, each lane is divided into lane segments of equal length according to a preset length, resulting in a set of lane segments for that precision. ; in, L This indicates the precision (length) of lane segments. U i Indicates the first i A set of lane segments with a precision of [number] u ij Indicates the first i The first precision j One lane segment, N i Indicates the first i The total number of lane segments in each precision level.

[0033] Simultaneously, the adjacency relationships between lane segments are recorded to obtain the adjacency relationship matrix at this level of precision: ; ; in, E i Indicates the first i The lane segment adjacency matrix at a given precision is: N i × N i A square matrix, matrix elements e ijk Indicates the first i At the precision of the first j The first lane segment and the first k Adjacency relationships between lane segments (e.g.) e ijk = 1 indicates adjacent, e ijk = 0 indicates that they are not adjacent. U i and E i Together constitute the first i Lane scene at a certain precision.

[0034] Optionally, in some embodiments of the present invention, the steps of obtaining the set of different lane segments traversed by the vehicle and multiple predicted paths specifically include: Historical features of vehicles are extracted using fully connected layers and historical trajectories.

[0035] Multi-precision lane features are extracted using fully connected layers and multi-precision lanes.

[0036] The probability of the lane segment traversed by the vehicle is obtained using the fully connected layer, the vehicle's historical features, and the multi-precision lane features.

[0037] Based on the probability, a set of different lane segments traversed by the vehicle is obtained through multiple samplings.

[0038] Based on multiple lane segment sets, a variety of predicted paths are obtained using a fully connected layer.

[0039] In this embodiment, the above process is specifically represented by the following formula: Use the fully connected layer and the vehicle's historical path X Extracting historical features of vehicles X : ; MLP stands for a fully connected layer of any number of layers.

[0040] Multi-precision lane segment features are extracted using a fully connected layer and all lane segments within the multi-precision lane. Each lane segment is input into a fully connected layer, and the output is a set of features for all multi-precision lane segments. U : ; in, v ij = MLP( u ij ) indicates the first i At the precision of the first j Lane segment features mapped by a fully connected layer.

[0041] Using the historical features X By combining the features of each multi-precision lane segment, a probability set of the lane segments traversed by the vehicle is obtained. P : ; Here, ⊙ represents calculating the product of each corresponding element in two feature vectors (Hadamard product), and the softmax function transforms the features into a set of probabilities. P , P ij Indicates that the vehicle has passed the first i At the precision of the first j The probability of a lane segment, N i Indicates the first i The total number of lane segments at each level of precision. L This indicates the quantity of lane accuracy.

[0042] Based on the probability set P Multiple samples are taken based on the probability magnitude to obtain the route taken by the vehicle. K 1 set of lane segments, each set containing K Two multi-precision lane segment features: ; Wherein, the sampling function represents a set of probabilities. P Sampling operation, K 1 indicates the number of sample groups. K 2 represents the number of multi-precision lane segment features sampled in each group, and the matrix elements are... z ij Indicates from U The first sample obtained from the middle sampling i The first group of samples j The multi-precision lane segment feature, the matrix of the first... i Column corresponds to a set of sampling results Z i .

[0043] Based on multiple sets of lane segments traversed by the vehicle, each sampled set of lane segments is input into a fully connected layer to generate a corresponding predicted path. K One prediction path: ; Optionally, in some embodiments of the present invention, the steps for obtaining the trained lane and path prediction model specifically include: Obtain the vehicle's actual path and the lanes it actually passed through.

[0044] By using the cross-entropy loss function and the mean squared error loss function, the error between the predicted path and the lanes it passes through and the actual path and the lanes it passes through is calculated, and a trained lane and path prediction model is obtained.

[0045] In this embodiment, the above process is specifically represented by the following formula: The total loss function consists of two parts: cross-entropy loss and mean squared error loss. ; Where CE represents the cross-entropy loss function and MSE represents the mean squared error loss function. P This represents the set of probabilities for the predicted lane segments. Q This represents the set of probabilities of actually passing lane segments. Y ∗ This represents the vehicle's actual path. The min function retrieves this path. K The minimum error between the predicted path and the true path in one possible path. This is achieved by minimizing the total loss value. loss Adjust the parameters of the fully connected layer to obtain the trained lane and path prediction model.

[0046] The cross-entropy loss function CE( P , Q The specific calculation method is as follows: ; in, p ij This indicates that the model predicts the vehicle will pass through the [number]th [location]. i At the precision of the first j The probability of a lane segment, q ij This indicates a real-world label, indicating when a vehicle actually passes by. i At the precision of the first j When the lane segment q ij = 1, otherwise q ij = 0, L The number of lane accuracy, N i For the firsti The total number of lane segments at a given level of precision.

[0047] The mean square error loss function MSE( Y i , Y ∗ The specific calculation method is as follows: ; in, Y i The model predicts the first i Path, Y i - Y ∗ This represents the positional deviation between the predicted path and the actual path at each coordinate point. Y i - Y ∗ ) 2 The AVG function represents the average of the squared deviations of all coordinate points, and thus the mean square error between the predicted path and the actual path.

[0048] Through the joint optimization of the cross-entropy loss function and the mean squared error loss function, the model learns driving decisions at the lane level and path generation at the coordinate level simultaneously during training.

[0049] This embodiment also provides a lane and path prediction model construction device, which includes: multi-precision lanes, a fully connected layer, a cross-entropy loss function, and a mean squared error loss function. The device includes: The first acquisition module is used to acquire the historical paths of all lanes and vehicles in the target scene. For its specific implementation, please refer to the relevant descriptions of the corresponding steps in the above method embodiments.

[0050] The first processing module is used to construct multi-precision lanes based on the obtained lanes. For its specific implementation, please refer to the relevant descriptions of the corresponding steps in the above method embodiments.

[0051] The second processing module is used to obtain a set of different lane segments traversed by the vehicle and multiple predicted paths based on the obtained multi-precision lane and vehicle historical paths. For specific implementation details, please refer to the relevant descriptions of the corresponding steps in the above method embodiments.

[0052] The second acquisition module is used to acquire the vehicle's actual path and the lanes it actually passed through. For its specific implementation, please refer to the relevant descriptions of the corresponding steps in the above method embodiments.

[0053] The third processing module is used to obtain a trained lane and path prediction model based on the obtained set of different lane segments traversed by the vehicle, multiple predicted paths, and the vehicle's actual path and the lane segments traversed.

[0054] Example 2 This embodiment provides a path prediction method, including: Obtain the historical paths of all lanes and vehicles in the target scene.

[0055] All lanes in the target scenario are input into the trained path prediction model obtained by the above path prediction model construction method to obtain multi-precision lanes.

[0056] The obtained multi-precision lanes and the vehicle's historical path are input into the trained path prediction model obtained using the above-described path prediction model construction method to obtain multiple predicted paths for the vehicle.

[0057] By following the steps above, the trained path prediction model is used to predict the vehicle's path, making the prediction results closer to the real path and improving the accuracy of the prediction results.

[0058] In this embodiment of the invention, the path prediction results of different path prediction methods are compared with the path prediction method (LPTP) provided in this embodiment of the invention. The experimental comparison results are shown in Table 1. Among them, ADE K The representation model provides K When predicting a path, the minimum distance between the predicted path and the actual path is used for ADE. K The smaller the value, the more accurate the prediction result; MR K The representation model provides K When predicting a path, the percentage of the predicted path located on the road is MR. K The larger the value, the more the prediction result matches the vehicle's driving behavior on the road.

[0059] Table 1 Based on the results in Table 1, it can be seen that the path prediction method LPTP proposed in this embodiment of the invention achieved the best results. (ADE5 and ADE) 10 Performance evaluation shows that the predicted path of the embodiments of the present invention is closer to the actual path, and the embodiments of the present invention can effectively improve the accuracy of the prediction results. MR5 and MR 10 Performance evaluation shows that the predicted path in this embodiment of the invention is more consistent with the driving behavior of vehicles on the road.

[0060] This embodiment also provides a path prediction device. The device includes: The third acquisition module is used to acquire the historical paths of all lanes and vehicles in the target scene. For its specific implementation, please refer to the relevant descriptions of the corresponding steps in the above method embodiments.

[0061] The fourth processing module is used to construct multi-precision lanes based on the obtained lanes. For its specific implementation, please refer to the relevant descriptions of the corresponding steps in the above method embodiments.

[0062] The fifth processing module is used to obtain multiple predicted paths for vehicles based on the obtained multi-precision lane and vehicle historical paths. For specific implementation details, please refer to the relevant descriptions of the corresponding steps in the above method embodiments.

[0063] The path prediction device described above is used to execute the path prediction method provided in the above method embodiments. Its implementation and principle are the same. For details, please refer to the relevant descriptions in the above method embodiments.

[0064] Through the collaborative efforts of the aforementioned components, the device can use a trained path prediction model to predict the vehicle's path, making the prediction results closer to the real path and improving the accuracy of the prediction results.

[0065] Example 3 like Figure 2 As shown, this embodiment also provides an electronic device. The electronic device may include a processor and a memory, wherein the processor and memory can be connected via a bus or other means.

[0066] The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0067] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the methods in the above method embodiments. The processor executes various functional applications and data processing by running the non-transitory software programs, instructions, and modules stored in the memory, thereby implementing the methods in the above method embodiments.

[0068] The memory may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, the memory may include high-speed random access memory and non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0069] One or more modules are stored in memory and, when executed by a processor, perform the methods described in the above method embodiments.

[0070] The specific details of the above-mentioned electronic device can be understood by referring to the relevant descriptions and effects in the above method embodiments, and will not be repeated here.

[0071] This invention also provides a computer-readable storage medium storing computer instructions for causing a computer to execute the methods described in the above-described method embodiments. The storage medium may be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium may also include combinations of the above types of memory.

[0072] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for constructing a lane and path prediction model, characterized in that, include: Obtain the historical paths of all lanes and vehicles in the target scene; Based on the described lanes, construct multi-precision lanes; Based on the multi-precision lanes, the historical paths, and the fully connected layer, a set of different lane segments traversed by the vehicle is obtained, and multiple predicted paths are obtained based on the different lane segment sets. Obtain the vehicle's actual path and the lanes it actually passed through; Based on the various predicted paths and the lane segments traversed by the vehicle, and the actual paths and lanes actually traversed, a trained lane and path prediction model is obtained using the cross-entropy loss function and the mean squared error loss function.

2. The method according to claim 1, characterized in that, Based on the lane, construct a multi-precision lane, including: Set at least two lane separation lengths; Each lane is segmented according to its respective lane segmentation length to obtain a set of lane segments corresponding to each lane segmentation length and the adjacency relationship between lane segments; The multi-precision lane is obtained by combining the set of lane segments corresponding to each lane segment length and their adjacent relationships.

3. The method according to claim 1, characterized in that, Based on the multi-precision lanes, the historical path, and the fully connected layer, a set of different lane segments traversed by the vehicle is obtained, including: Based on the historical path and the fully connected layer, extract the historical features of the vehicle; Based on the multi-precision lane and the fully connected layer, extract the multi-precision lane features; Based on the historical features, the multi-precision lane features, and the fully connected layer, the probability of the vehicle passing through each lane segment is obtained; Multiple samples are taken based on the given probability to obtain the set of different lane segments.

4. The method according to claim 3, characterized in that, Based on the historical path and the fully connected layer, historical features of the vehicle are extracted, including: The historical path is input into the fully connected layer, and the historical features are output after mapping by the fully connected layer.

5. The method according to claim 3, characterized in that, Based on the multi-precision lanes and the fully connected layer, multi-precision lane features are extracted, including: Each lane segment in the multi-precision lane is input into the fully connected layer, and the multi-precision lane features corresponding to each lane segment are output through mapping by the fully connected layer.

6. The method according to claim 4, characterized in that, Based on the historical features, the multi-precision lane features, and the fully connected layer, the probability of the vehicle traversing each lane segment is obtained, including: Calculate the product of the historical features and each corresponding element in the multi-precision lane features to obtain the fused features; The fused features are input into the fully connected layer, mapped by the fully connected layer, and then converted into the probabilities by the softmax function.

7. The method according to claim 6, characterized in that, Based on the stated probability, multiple samples are taken to obtain the set of different lane segments, including: Based on the probability, multiple sets of lane segments are obtained by sampling, and each set of lane segments contains at least one lane segment.

8. The method according to claim 1, characterized in that, Based on the various predicted paths and the lane segments traversed by the vehicles, and comparing them with the actual paths and lanes actually traversed, a trained lane and path prediction model is obtained using the cross-entropy loss function and the mean squared error loss function, including: The first loss value is calculated using the cross-entropy loss function based on the probability of the vehicle passing through each lane segment and the lane actually passed. Based on the multiple predicted paths and the actual path, the mean square error between each predicted path and the actual path is calculated, and the minimum mean square error is selected as the second loss value. The total loss value is obtained based on the first loss value and the second loss value; The parameters of the fully connected layer are adjusted based on the total loss value to obtain the trained lane and path prediction model.

9. A path prediction method, characterized in that, include: Obtain the historical paths of all lanes and vehicles in the target scene; Set at least two lane segmentation lengths, segment each lane according to each lane segmentation length, obtain the lane segment set corresponding to each lane segmentation length and the adjacency relationship between lane segments, combine the lane segment set corresponding to each lane segmentation length and the adjacency relationship to obtain a multi-precision lane; Historical features of the vehicle are extracted based on the historical path and the fully connected layer. Multi-precision lane features are extracted based on the multi-precision lane and the fully connected layer. The probability of the vehicle passing through each lane segment is obtained based on the historical features, the multi-precision lane features, and the fully connected layer. Multiple samplings are performed based on the stated probability to obtain multiple lane segment sets, and corresponding prediction paths are generated based on each lane segment set, thereby obtaining multiple prediction paths for the vehicle.

10. An electronic device comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, characterized in that, When the processor executes the computing program, it implements the method of any one of claims 1-9.

Images