Automatic parking method and device, storage medium and computer device

Abstract

The application provides an automatic parking method and device, a storage medium and a computer device. The method comprises the following steps: performing scene recognition by using a scene recognition model to determine a target parking scene; selecting a parking space selection model corresponding to the target parking scene from a plurality of parking space selection models as a target parking space selection model according to the target parking scene, and selecting a trajectory planning model corresponding to the target parking scene from a plurality of trajectory planning models as a target trajectory planning model; selecting a target parking space in a current road environment by using the target parking space selection model; planning a target trajectory by using the target trajectory planning model, the target trajectory being used to indicate that a vehicle moves from a current position to the target parking space; and controlling the vehicle to park according to a parking mode corresponding to the target parking scene and the target trajectory. The method can be adapted to various complex scenes, improves flexibility, and the used model has optimal parking space selection ability and path planning ability in the corresponding scene, and improves accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic driving, and particularly relates to an automatic parking method and device, a storage medium and a computer device. BACKGROUND

[0002] With continuous iteration of automatic driving technology, automatic driving technology gradually enters people's life, greatly facilitating people's life, and people's demand for automatic driving is increasingly urgent. For automatic driving technology, the vehicle automatically perceives the road surface condition and automatically plans a trajectory to send the passenger to the destination. When reaching the vicinity of the destination, automatic parking is needed. The traditional automatic parking method has the problem of unreasonable selection of parking spaces and trajectories. SUMMARY

[0003] The present application aims to at least solve one of the above technical defects, in particular, the technical defect of unreasonable selection of parking spaces and trajectories in the prior art.

[0004] In a first aspect, the present application provides an automatic parking method, comprising: performing scene recognition by using a scene recognition model to determine a target parking scene; selecting, according to the target parking scene, a parking space selection model corresponding to the target parking scene from a plurality of parking space selection models as a target parking space selection model, and selecting a trajectory planning model corresponding to the target parking scene from a plurality of trajectory planning models as a target trajectory planning model; selecting a target parking space in a current road environment by using the target parking space selection model; planning a target trajectory by using the target trajectory planning model, the target trajectory being used to indicate movement of a vehicle from a current position to the target parking space; and controlling the vehicle to park according to a parking mode corresponding to the target parking scene and the target trajectory.

[0005] In one of the embodiments, the controlling the vehicle to park according to the parking mode corresponding to the target parking scene and the target trajectory comprises: controlling the vehicle to park according to the parking mode corresponding to the target parking scene and the target trajectory, and when a time length of parking in the current target trajectory reaches a preset time length, returning to the step of selecting the target parking space in the current road environment by using the target parking space selection model to continue execution, so as to update the target parking space and the target trajectory until the vehicle reaches the updated target parking space.

[0006] In one of the embodiments, the performing scene recognition by using the scene recognition model to determine the target parking scene comprises: in a case where a distance between the vehicle and an ideal parking point is less than a preset distance, acquiring a roadbed structure map around the vehicle; and inputting the roadbed structure map into the scene recognition model to perform scene recognition and determine the target parking scene.

[0007] In one of the embodiments, the training process of the scene recognition model comprises: obtaining a first training set; the first training set comprises annotated roadbed structure maps corresponding to different parking scenes, and the annotation of the annotated roadbed structure map is used to indicate the parking scene corresponding to the annotated roadbed structure map; inputting the first training set into an initial scene recognition model to obtain scene recognition results of each annotated roadbed structure map; adjusting the initial scene recognition model according to the difference between each scene recognition result and the corresponding annotation to obtain a trained scene recognition model.

[0008] In one of the embodiments, the training process of the parking space selection model corresponding to any parking scene comprises: obtaining a second training set; the second training set comprises a plurality of first training maps of the parking scene corresponding to the parking space selection model, and the annotation of the first training map is used to indicate the standard parking space; inputting the second training set into an initial parking space selection model to obtain predicted parking spaces corresponding to each first training map; adjusting the initial parking space selection model according to the difference between the predicted parking space corresponding to each first training map and the standard parking space to obtain a trained parking space selection model.

[0009] In one of the embodiments, obtaining the second training set comprises: obtaining a plurality of different environment map packages; each environment map package comprises a plurality of continuous environment maps for parking in the parking scene corresponding to the parking space selection model; for any one environment map package, extracting one environment map from the environment map package every preset frame interval, and obtaining a first training map according to the extracted environment map; annotating each first training map to obtain the second training set.

[0010] In one of the embodiments, obtaining the first training map according to the extracted environment map comprises: placing obstacles with different congestion degrees in the extracted environment map to obtain a plurality of first training maps corresponding to each extracted environment map.

[0011] In one of the embodiments, adjusting the initial parking space selection model according to the difference between the predicted parking space corresponding to each first training map and the annotation of each first training map comprises: adjusting the initial parking space selection model according to the difference between the predicted parking space corresponding to each first training map and the annotation of each first training map and the prediction stability; the prediction stability is used to reflect the fluctuation between the predicted parking spaces corresponding to each first training map belonging to the same environment map package.

[0012] In one of the embodiments, the training process of the trajectory planning model corresponding to any one parking scenario includes: obtaining a third training set; the third training set includes a plurality of second training maps of the parking scenario corresponding to the trajectory planning model, and the label of the second training map is used to indicate a standard trajectory; inputting the third training set into an initial trajectory planning model to obtain a predicted trajectory corresponding to each second training map; adjusting the initial trajectory planning model according to the difference between the predicted trajectory corresponding to each second training map and the standard trajectory to obtain a trained trajectory planning model.

[0013] In a second aspect, the embodiments of the present application provide an automatic parking device, including: a scene determination module, configured to determine a target parking scenario by using a scene recognition model; a model selection module, configured to select a parking space selection model corresponding to the target parking scenario as a target parking space selection model from a plurality of parking space selection models according to the target parking scenario, and select a trajectory planning model corresponding to the target parking scenario as a target trajectory planning model from a plurality of trajectory planning models; a parking space determination module, configured to select a target parking space in a current road environment by using the target parking space selection model; a trajectory determination module, configured to plan a target trajectory by using the target trajectory planning model, the target trajectory being used to indicate a movement of a vehicle from a current position to the target parking space; and a parking module, configured to control the vehicle to park according to a parking mode corresponding to the target parking scenario and the target trajectory.

[0014] In a third aspect, the embodiments of the present application provide a computer device, including one or more processors, and a memory, and the memory stores computer readable instructions, and the computer readable instructions are executed by the one or more processors to perform the steps of the automatic parking method in any one of the above embodiments.

[0015] In a fourth aspect, the embodiments of the present application provide a storage medium, and the storage medium stores computer readable instructions, and the computer readable instructions are executed by one or more processors to make the one or more processors perform the steps of the automatic parking method in any one of the above embodiments.

[0016] From the above technical solutions, the embodiments of the present application have the following advantages:

[0017] Based on any one of the above embodiments, the road environment in which the vehicle is located is considered, the parking scenario in which the vehicle is located is recognized, the parking scenario is taken as a target parking scenario, a target parking space is selected by using a target parking space selection model adapted to the target parking scenario, a target trajectory is planned by using a target trajectory planning model, and finally a parking mode corresponding to the target parking scenario is selected to control the vehicle to park into the target parking space along the target trajectory. The method can be adapted to automatic parking in various complex scenarios, improves the flexibility of automatic parking, and the models used have optimal parking space selection ability and path planning ability in the corresponding scenarios, thereby improving the accuracy of automatic parking. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0019] Figure 1 The flowchart of the automatic parking method provided by the embodiments of the present application is shown in the figure.

[0020] Figure 2 The training flowchart of the scene recognition model provided by the embodiments of the present application is shown in the figure.

[0021] Figure 3 The training flowchart of the parking space selection model provided by the embodiments of the present application is shown in the figure.

[0022] Figure 4 The flowchart of obtaining the second training set provided by the embodiments of the present application is shown in the figure.

[0023] Figure 5 The training flowchart of the trajectory planning model provided by the embodiments of the present application is shown in the figure.

[0024] Figure 6 The module schematic diagram of the automatic parking device provided by the embodiments of the present application is shown in the figure.

[0025] Figure 7 The internal structure diagram of the computer device provided by the embodiments of the present application is shown in the figure. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0027] The present application provides an automatic parking method, which can be used as a supplement to manual driving or as part of full automatic driving. That is, the automatic parking algorithm can be used for parking when the driver manually drives the vehicle to the vicinity of the parking point, solving the problem of parking difficulty for some drivers. It can also be used in the scenario of full automatic driving, that is, the user selects the destination for the vehicle by interacting with the vehicle, and the vehicle drives from the starting point to the vicinity of the destination according to the automatic driving algorithm. When reaching the vicinity of the destination, the automatic parking method is used for parking. Please refer to Figure 1The automatic parking method comprises steps S102 to S110.

[0028] S102, scene recognition is performed by using a scene recognition model to determine a target parking scene.

[0029] It can be understood that the parking scene herein can indicate the characteristics of the surrounding road structure when parking, that is, if the surrounding road structure is similar when parking, it will be classified into the same parking scene. In the field of automatic driving technology, the vehicle will detect the current road environment through detection devices such as radar and / or camera, thereby collecting road network information to instruct the vehicle to automatically drive. The scene recognition model can recognize the parking scene of the vehicle when parking according to all or part of the information in the road network information after training, and take the scene as the target parking scene. Common parking scenes include ordinary curb parking scene, taxi parking scene, bus parking scene, side parking scene, reversing into garage scene, special section (turning, intersection, construction, etc.) parking scene, etc. The more detailed the parking scene divided by the scene recognition model during training, the better the subsequent parking space and trajectory recognition effect, but it may lead to an increase in training cost. Therefore, scene division can be performed according to actual conditions when training the scene recognition model.

[0030] S104, according to the target parking scene, a parking space selection model corresponding to the target parking scene is selected from a plurality of parking space selection models as a parking space selection target model, and a trajectory planning model corresponding to the target parking scene is selected from a plurality of trajectory planning models as a trajectory planning target model.

[0031] It can be understood that the disclosure adopts the trained parking space selection module to select a parking space for the vehicle, and adopts the trained trajectory planning model to select a trajectory for the vehicle to move from the current position to the parking space. The parking space selection module and the trajectory planning model are also based on all or part of the information in the road network information to select a parking space and plan a trajectory. Different parking scenes mean that the road environment where the vehicle is located is different, and the selection of the parking space and the planning of the parking path are related to the road environment.

[0032] Based on this, the disclosure trains a one-to-one corresponding parking space selection model and trajectory planning model for each parking scene divided in the scene recognition model. That is, for any parking scene, only the parking space selection model and the trajectory planning model under the parking scene are trained based on the collected road network information, to improve their parking space selection ability and trajectory planning ability under the parking scene, without paying attention to their parking space selection ability and trajectory planning ability under other parking scenes, so that the finally trained parking space selection model and trajectory planning model have the optimal parking space selection ability and trajectory planning ability under the corresponding parking scene.

[0033] S106, selecting a target parking space in the current road environment by using a parking space selection target model.

[0034] Based on the above description, the parking space selection target model is a parking space selection model with the best parking space selection capability in the target parking scenario, and the target parking space is the parking space recognition result obtained after inputting the information required by the parking space selection target model.

[0035] S108, planning a target trajectory by using a trajectory planning target model, the target trajectory being used to indicate the movement of the vehicle from the current position to the target parking space.

[0036] Similarly, the trajectory planning target model is a trajectory planning model with the best trajectory planning capability in the target parking scenario, and the target trajectory is the trajectory planning result obtained after inputting the information required by the parking space selection target model.

[0037] S110, controlling the vehicle to park according to the parking mode corresponding to the target parking scenario and the target trajectory.

[0038] It can be understood that different control modes should be adopted to control the vehicle to park in different road environments. For example, in the reverse parking scenario and the side parking scenario, the vehicle is generally required to be parked near the parking space, and then the vehicle body direction is adjusted to be perpendicular to the parking space before starting to reverse, and the vehicle body direction is gradually adjusted to be parallel to the parking space during the reversing process. In the side parking scenario, the vehicle is parked near the parking space, and then the vehicle body direction is adjusted to be parallel to the parking space before starting to reverse, and the vehicle body direction is gradually increased to be parallel to the parking space during the reversing process. When the vehicle reaches a certain position, the direction is adjusted to be parallel to the parking space, and finally the vehicle body direction is adjusted to be parallel to the parking space. For example, in the taxi parking scenario and the bus parking scenario, the taxi has a large degree of freedom when parking, and can be parked in any non-illegal parking area near the terminal. However, the bus must be driven to the bus station when parking, and if the bus station is occupied, the bus must wait in line to enter the station.

[0039] Therefore, for each parking scenario, a parking algorithm suitable for the parking scenario should be adopted to control the vehicle to park. That is, after selecting the target parking scenario, the vehicle is controlled to park in the target parking space along the target trajectory planned by the trajectory planning target model by using the parking mode suitable for the target parking scenario.

[0040] Based on the automatic parking method in the embodiment, the parking scene where the vehicle is located is identified considering the road environment where the vehicle is located, the target parking scene is taken as a target parking scene, the target parking space is selected by using a parking space selection target model adapted to the target parking scene, and the target trajectory is planned by using a trajectory planning target model, and finally the target parking scene corresponding parking mode is selected to control the vehicle to park into the target parking space along the target trajectory. The method can be adapted to automatic parking in various complex scenes, improves the flexibility of automatic parking, and the model used has optimal parking space selection ability and path planning ability in the corresponding scene, improving the accuracy of automatic parking.

[0041] In one embodiment, the vehicle is parked according to the target parking scene corresponding parking mode and the target trajectory, including: controlling the vehicle to park according to the target parking scene corresponding parking mode and the target trajectory, and when the time length of parking in the current target trajectory reaches a preset time length, returning to the step of selecting the target parking space in the current road environment by using the parking space selection target model to continue execution, to update the target parking space and the target trajectory, until the vehicle reaches the updated target parking space.

[0042] It can be understood that road traffic is often a dynamic system, and changes in road conditions can occur at any time. Even in a closed parking lot, a vehicle that interferes with parking may suddenly appear, such as the target parking space being occupied by a person during the automatic parking process, not to mention various emergencies in an open traffic environment. Therefore, the automatic parking in the embodiment is a dynamic planning process. After selecting a target parking space and planning a target trajectory to the target parking space, parking is performed according to the target parking space and the target trajectory. When the time length of parking in the target parking space and the target trajectory reaches a preset time length, the road environment where the vehicle is located may have changed, such as the current target parking space or target trajectory being blocked, and then the target parking space is updated by using the parking space selection target model according to the current road environment. The target trajectory is updated by using the trajectory planning target model according to the updated target parking space. The vehicle can park according to the updated target parking space and target trajectory. In summary, the target parking space and the target trajectory can be adjusted according to the change of the road environment during parking, until the vehicle reaches the final target parking space. The embodiment ensures that the vehicle can adapt to changes in the road environment, improving the environmental adaptability of automatic parking.

[0043] In one embodiment, the scene recognition model is used for scene recognition to determine the target parking scene, including:

[0044] 1) In the case where the distance between the vehicle and the ideal parking point is less than a preset distance, the roadbed structure map around the vehicle is obtained.

[0045] It can be understood that the ideal parking point can be selected by the user after interacting with the vehicle when full automatic driving, and the parking space selected by the related algorithm without considering the actual road environment near the destination. It can also be a parking space selected by the user or recognized by the related algorithm when the user drives manually to the vicinity of the destination. The distance between the vehicle and the ideal parking point is less than the preset distance, which means that automatic parking can be started. In this embodiment, the division of the parking scene should be based on at least the roadbed structure diagram in the road network information. The roadbed structure diagram is composed of road elements and can reflect the structure diagram of various road characteristics such as road pattern and shape. The acquisition method of the roadbed structure diagram is mature in the field of automatic driving, including but not limited to discretizing road elements into points from the high-precision map perceived by the vehicle, and then distributing them on a two-dimensional grid network to obtain the roadbed structure diagram.

[0046] 2) Input the roadbed structure diagram into the scene recognition model to recognize the scene and determine the target parking scene.

[0047] The input of the scene recognition model should at least include the roadbed structure diagram, but some embodiments divide the parking scene more finely, such as considering lane line, traffic light position and other information. Therefore, when training the scene recognition model, information other than the roadbed structure diagram should also be considered, and information other than the roadbed structure diagram should be extracted from the road network information and input into the scene recognition model for scene recognition.

[0048] In one of the embodiments, please refer to Figure 2 The training process of the scene recognition model includes steps S202 to S206.

[0049] S202, obtaining a first training set.

[0050] The first training set includes labeled roadbed structure diagrams corresponding to different parking scenes. The label of the labeled roadbed structure diagram is used to indicate the parking scene corresponding to the labeled roadbed structure diagram. It can be understood that each of the plurality of different parking scenes divided according to the need will correspond to a plurality of labeled roadbed structure diagrams. The label of the labeled roadbed structure diagram indicates which parking scene the labeled roadbed structure diagram belongs to. According to the need of scene division, a detection device can be installed on a vehicle used daily in different parking scenes, and roadbed structure diagrams in each parking scene can be collected through the detection device during normal driving, and then the roadbed structure diagrams are labeled to obtain the labeled roadbed structure diagrams. It can also be extracted from the road network information collected in various parking scenes during automatic driving test, display and other processes.

[0051] S204, inputting the first training set into an initial scene recognition model to obtain the scene recognition result of each labeled roadbed structure diagram.

[0052] The initial scene recognition model can recognize scenes of each labeled roadbed structure diagram, and a scene recognition result is a parking scene to which each labeled roadbed structure diagram belongs, which is predicted by the initial scene recognition model.

[0053] In S206, the initial scene recognition model is adjusted according to differences between each scene recognition result and a corresponding label, to obtain a trained scene recognition model.

[0054] Specifically, each labeled roadbed structure diagram corresponds to a scene recognition result and a label, the scene recognition result represents a predicted parking scene, and the label represents an actual parking scene. The difference between the prediction and the actual can be used to determine the recognition ability of the initial scene recognition model for the scene, and the smaller the difference is, the stronger the recognition ability of the initial scene recognition model is. Therefore, the initial scene recognition model is adjusted to reduce the difference between the scene recognition result and the corresponding label as the goal, until the initial scene recognition model meets a first training end condition, and then a trained scene recognition model is obtained. A first loss function can be constructed according to the difference between the output scene recognition result and the corresponding label, and when the first loss function is less than a first end threshold, it is determined that the scene recognition model meets the first training end condition. In addition, the initial scene recognition model meeting the first loss function being less than the first end threshold may only mean that the initial scene recognition model has strong recognition ability for similar roadbed structure diagrams in the first training set. A first verification set unrelated to the first training set can be obtained in the manner of obtaining the first training set, to verify whether the initial scene recognition model needs to be further adjusted.

[0055] In one of the embodiments, referring to Figure 3 The training process of the parking space selection model corresponding to any one of the parking scenes includes steps S302 to S306.

[0056] In S302, a second training set is obtained.

[0057] The second training set includes a plurality of first training maps of the parking scene corresponding to the parking space selection model, and the label of the first training map is used to indicate a standard parking space. It can be understood that for each parking space selection model, its ability to select a parking space in a road environment related to the corresponding parking scene needs to be improved. Therefore, the first training map can reflect the road environment in the parking scene corresponding to the parking space selection model, and at least the roadbed structure near the vehicle should be included in the first training map. According to the needs of scene division, lane lines, traffic light positions, obstacles, etc. can also be added to the first training map. The standard parking space is a relatively appropriate parking position, and the label of the first training map can indicate the standard parking space.

[0058] In S304, the second training set is input into the initial parking space selection model to obtain a predicted parking space corresponding to each first training map.

[0059] The initial parking space selection model can perform parking space selection on each first training map, and the selection result is the predicted parking space of the initial parking space selection model on each first training map.

[0060] S306, according to the difference between the predicted parking space corresponding to each first training map and the standard parking space, adjusting the initial parking space selection model to obtain a trained parking space selection model.

[0061] Specifically, each first training map corresponds to a standard parking space and a predicted parking space. According to the difference between the standard parking space and the predicted parking space, the ability of the initial parking space selection model to select parking spaces in the corresponding parking scene can be determined. The smaller the difference, the stronger the selection ability of the initial parking space selection model. Therefore, the goal is to narrow the difference between the labeled parking space and the predicted parking space in the same training map, and adjust the initial parking space selection model until the initial parking space selection model meets the second training end condition, and then obtain the trained parking space selection model. The second loss function can be constructed according to the difference between the labeled parking space and the predicted parking space. When the second loss function is less than the second end threshold, it is determined that the initial parking space selection model meets the second training end condition. In addition, the initial parking space selection model meeting the second loss function less than the second end threshold may only mean that the initial parking space selection model has strong selection ability for the first training map in the second training set. A plurality of new road environments corresponding to the parking scene can be collected as a second verification set to verify whether the initial parking space selection model needs to be further adjusted.

[0062] In one embodiment, please refer to Figure 4 The specific way of obtaining the second training set includes steps S402 to S406.

[0063] S402, obtaining a plurality of different environment map packages.

[0064] Each environment map package includes a plurality of continuous environment maps for parking in the parking scene corresponding to the parking space selection model. It can be understood that during the parking process of the vehicle in the parking scene corresponding to the parking space selection model, the road environment around the vehicle is continuously collected to obtain the environment map package. The parking scene is divided according to whether the road environment has some specific common points, and the road environment features not considered during the division may have some differences. Different environment map packages refer to environment map packages with different road environment features that have not been considered, which can enrich the adaptability of parking space selection. In addition, since each environment map in the same environment map package contains the entire parking process, it can reflect the selection of parking spaces at different distances from the final parking position, so that the initial parking space selection model can learn the knowledge of parking space selection at each parking stage.

[0065] S404, for any one environment map package, extract one frame of environment map from the environment map package every interval preset frame number, and obtain a first training map according to the extracted environment map.

[0066] Since the environment map package is continuously recorded, the difference between the frames of environment map close to each other is not big, and the learning value of the relatively similar data is not big for the model. In order to improve the training efficiency, one frame of environment map is extracted every interval preset frame number. The extracted environment map can be directly used as the first training map. In order to expand the scale of the data set, the extracted environment map can also be deformed to obtain multiple first training maps.

[0067] For example, the obstacle that hinders parking in the road environment is an important factor affecting the selection of parking space. In some embodiments, by placing obstacles with different congestion levels in the extracted environment map, multiple first training maps corresponding to each extracted environment map are obtained. Different congestion levels mean that the number of obstacles in each first training map is different. The obstacles can be actual obstacles around the vehicle when the frame of environment map is collected, such as a vehicle driving slowly or a vehicle parked on the roadside. By changing the number and position of these obstacles, multiple first training maps are obtained. It can also be to place simulated obstacles with different numbers and positions in the extracted environment map. This way makes the model learn how to select a parking space under the influence of obstacles with different congestion levels.

[0068] S406, labeling each first training map to obtain a second training set.

[0069] When labeling, it can rely on manual selection on the first training map. It can also rely on a parking space selection algorithm with unguaranteed accuracy to roughly select a rough parking space, and then manually correct the rough parking space to obtain a standard parking space.

[0070] In one embodiment, the initial parking space selection model is adjusted according to the difference between the predicted parking space corresponding to each first training map and the annotation of each first training map, including: adjusting the initial parking space selection model according to the difference between the predicted parking space corresponding to each first training map and the annotation of each first training map and the prediction stability. The prediction stability is used to reflect the fluctuation between the predicted parking spaces corresponding to each first training map belonging to the same environment map package.

[0071] It can be understood that the parking process dynamically adjusts the target parking space. For the embodiment of the target parking space, the newly selected target parking space should not deviate too much from the previous target parking space. That is, the prediction of the parking space selection model in the parking process should remain stable, and each first training map of the same environment map package reflects the road environment at each stage in the same parking process. According to the fluctuations between the predicted parking spaces selected by the initial parking space selection model on each first training map of the same environment map package, it can be determined whether the initial parking space selection model has the ability to stably predict. Therefore, when constructing the second loss function, a difference term related to the prediction stability can be added to comprehensively adjust the initial parking space selection model based on the prediction rationality (i.e., the difference between the predicted parking space and the corresponding labeled parking space) and the prediction stability.

[0072] In one of the embodiments, referring to Figure 5 , the training process of the trajectory planning model corresponding to any one parking scene includes steps S502 to S506.

[0073] S502, obtaining a third training set.

[0074] The third training set includes a plurality of second training maps of the parking scene corresponding to the trajectory planning model, and the label of the second training map is used to indicate a standard trajectory. It can be understood that for each trajectory planning model, its ability to plan a trajectory in the road environment related to the corresponding parking scene needs to be improved. Therefore, the second training map can reflect the road environment in the parking scene corresponding to the trajectory planning model, and the second training map should at least include the roadbed structure near the vehicle and the position of the standard parking space. According to the needs of scene division, lane lines, traffic light positions, obstacles, etc. can also be added to the second training map. The standard trajectory indicates the route of the vehicle moving from the current position to the standard parking space. The second training map can be independent of the first training map. For example, a plurality of parking trajectories, final parking positions, and road environments during actual parking are obtained, the final parking position is taken as the standard parking space, the second training map is obtained according to the road environment, and the second training map is labeled according to the parking trajectory and the standard parking space of the same group. It can also be obtained by further marking on the first training map. That is, the first training map has been labeled with a standard parking space, and a trajectory from the current position of the vehicle to the standard parking space can be manually drawn based on this.

[0075] S504, inputting the third training set into the initial trajectory planning model to obtain a predicted trajectory corresponding to each second training map.

[0076] The initial trajectory planning model can plan a trajectory on each second training map, and the trajectory result is a predicted trajectory of the initial parking space selection model in the second training map, predicting the movement of the vehicle from the current position to the standard parking space.

[0077] S506, Based on the differences between the predicted trajectory and the standard trajectory corresponding to each second training map, adjust the initial trajectory planning model to obtain the trained trajectory planning model.

[0078] Specifically, each second training map corresponds to a standard trajectory and a predicted trajectory. The difference between the standard trajectory and the predicted trajectory determines the initial trajectory planning model's ability to plan trajectories in the corresponding parking scenario; the smaller the difference, the stronger the initial trajectory planning model's selection ability. Therefore, with the goal of reducing the difference between the standard trajectory and the predicted trajectory in the same training map, the initial trajectory planning model is adjusted until it meets the third training termination condition, resulting in a trained trajectory planning model. A third loss function can be constructed based on the difference between the standard trajectory and the predicted trajectory. When the third loss function is less than the third termination threshold, the initial trajectory planning model is considered to have met the third training termination condition. However, the initial trajectory planning model meeting the third termination threshold may only mean that it has a strong planning ability on the second training maps in the third training set. Multiple new road environments can be collected in the corresponding parking scenario, and standard parking spaces can be selected as the third validation set to verify whether the initial trajectory planning model needs further adjustment.

[0079] Secondly, this application provides an automatic parking device, please refer to [link to relevant documentation]. Figure 6 The system includes a scene determination module 210, a model selection module 220, a parking space determination module 230, a trajectory determination module 240, and a parking module 250. The scene determination module uses a scene recognition model to identify the target parking scene. The model selection module selects a parking space selection model corresponding to the target parking scene from multiple parking space selection models, and selects a trajectory planning model corresponding to the target parking scene from multiple trajectory planning models. The parking space determination module uses the parking space selection target model to select a target parking space in the current road environment. The trajectory determination module uses the trajectory planning target model to plan a target trajectory, which indicates the vehicle's movement from its current position to the target parking space. The parking module controls the vehicle to park according to the parking method corresponding to the target parking scene and the target trajectory.

[0080] Specific limitations regarding the automatic parking device can be found in the limitations of the automatic parking method described above, and will not be repeated here. Each module in the aforementioned automatic parking device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module. It should be noted that the module division in this embodiment is illustrative and only represents a logical functional division; in actual implementation, there may be other division methods.

[0081] Thirdly, embodiments of this application provide a computer device including one or more processors and a memory storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, they perform the steps of the automatic parking method in any of the above embodiments.

[0082] Fourthly, embodiments of this application provide a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the automatic parking method in any of the above embodiments.

[0083] Indicatively, such as Figure 7 As shown, Figure 7 This is a schematic diagram of the internal structure of a computer device 300 provided in an embodiment of this application. The computer device 300 can be provided as a server. (Refer to...) Figure 7 The computer device 300 includes a processing component 302, which further includes one or more processors, and memory resources represented by memory 301 for storing instructions, such as application programs, that can be executed by the processing component 302. The application programs stored in memory 301 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 302 is configured to execute instructions to perform the text recognition method of any of the above embodiments.

[0084] The computer device 300 may also include a power supply component 303 configured to perform power management of the computer device 300, a wired or wireless network interface 304 configured to connect the computer device 300 to a network, and an input / output (I / O) interface 305. The computer device 300 may operate on an operating system stored in memory 301, such as Windows Server™, Mac OS X™, Unix™, Linux™, Free BSD™, or similar.

[0085] Those skilled in the art will understand that Figure 7The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0086] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0087] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0088] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An automatic parking method characterized by, The method comprises the following steps: scene recognition is performed by using a scene recognition model to determine a target parking scene; the scene recognition model is an identification model that can recognize a parking scene of a vehicle when parking according to all or part of information in road network information after training, and the parking scene is taken as a target parking scene; a target parking space selection model corresponding to the target parking scene is selected from a plurality of parking space selection models according to the target parking scene, and a target trajectory planning model corresponding to the target parking scene is selected from a plurality of trajectory planning models; the target parking space selection model is a parking space selection model that has the best parking space selection capability under the target parking scene; a target parking space is selected in a current road environment by using the target parking space selection model; the target parking space is a parking space recognition result obtained after information required by the target parking space selection model is input; a target trajectory is planned by using the target trajectory planning model, and the target trajectory is used to indicate movement of the vehicle from a current position to the target parking space; the vehicle is parked according to a parking mode corresponding to the target parking scene and the target trajectory.

2. The method of claim 1, wherein, The step of parking the vehicle according to the parking mode corresponding to the target parking scene and the target trajectory comprises the following steps: the vehicle is parked according to the parking mode corresponding to the target parking scene and the target trajectory, and when a parking time of the vehicle according to the target trajectory reaches a preset time, the step of selecting the target parking space in the current road environment by using the target parking space selection model is executed again to update the target parking space and the target trajectory until the vehicle reaches the updated target parking space.

3. The method of claim 1, wherein, The step of performing scene recognition by using the scene recognition model to determine the target parking scene comprises the following steps: when a distance between the vehicle and an ideal parking point is less than a preset distance, a roadbed structure map around the vehicle is obtained; the roadbed structure map is input into the scene recognition model to perform scene recognition and determine the target parking scene.

4. The method of claim 3, wherein, The training process of the scene recognition model comprises the following steps: a first training set is obtained; the first training set comprises labeled roadbed structure maps corresponding to different parking scenes, and the labels of the labeled roadbed structure maps are used to indicate the parking scenes corresponding to the labeled roadbed structure maps; the first training set is input into an initial scene recognition model to obtain scene recognition results of the labeled roadbed structure maps; the initial scene recognition model is adjusted according to differences between the scene recognition results and the corresponding labels to obtain the trained scene recognition model.

5. The method of claim 1, wherein, The training process of the parking space selection model corresponding to any one of the parking scenes comprises the following steps: a second training set is obtained; the second training set comprises a plurality of first training maps of the parking scene corresponding to the parking space selection model, and labels of the first training maps are used to indicate standard parking spaces; the second training set is input into an initial parking space selection model to obtain predicted parking spaces corresponding to the first training maps. According to differences between the predicted parking space corresponding to each of the first training maps and the standard parking space, the initial parking space selection model is adjusted to obtain a trained parking space selection model.

6. The method of claim 5, wherein, The second training set is obtained by: Obtaining a plurality of different environment map packages; each of the environment map packages includes a plurality of continuous environment maps for parking in the parking scene corresponding to the parking space selection model; For any one of the environment map packages, a frame of the environment map is extracted from the environment map package every preset number of frames, and the first training map is obtained according to the extracted environment map; Each of the first training maps is labeled to obtain the second training set.

7. The method of claim 6, wherein, The first training map is obtained according to the extracted environment map, including: Different congestion levels of obstacles are placed in the extracted environment map to obtain a plurality of first training maps corresponding to each of the extracted environment maps.

8. The method of claim 6, wherein, According to the differences between the predicted parking space corresponding to each of the first training maps and the label of each of the first training maps, the initial parking space selection model is adjusted, including: According to the differences between the predicted parking space corresponding to each of the first training maps and the label of each of the first training maps, and the prediction stability, the initial parking space selection model is adjusted; the prediction stability is used to reflect the fluctuation between the predicted parking spaces corresponding to each of the first training maps belonging to the same environment map package.

9. The method of claim 1, wherein, The training process of the trajectory planning model corresponding to any one of the parking scenes includes: Obtaining a third training set; the third training set includes a plurality of second training maps of the parking scene corresponding to the trajectory planning model, and the label of the second training map is used to indicate a standard trajectory; The third training set is input into an initial trajectory planning model to obtain a predicted trajectory corresponding to each of the second training maps; According to the differences between the predicted trajectory corresponding to each of the second training maps and the standard trajectory, the initial trajectory planning model is adjusted to obtain a trained trajectory planning model.

10. An automatic parking apparatus characterized by comprising: Including: A scene determination module is configured to determine a target parking scene by using a scene recognition model; The scene recognition model is a trained model that can recognize the parking scene of a vehicle when parking according to all or part of the road network information, and uses the parking scene as the target parking scene; A model selection module is configured to select, according to the target parking scene, a parking space selection model corresponding to the target parking scene from a plurality of parking space selection models as a parking space selection target model, and select a trajectory planning model corresponding to the target parking scene from a plurality of trajectory planning models as a trajectory planning target model; the parking space selection target model is a parking space selection model with the best parking space selection capability in the target parking scene; A parking space determination module is configured to select a target parking space in the current road environment by using the parking space selection target model; The target parking space is a parking space recognition result obtained by inputting information required by the parking space selection target model; a trajectory determination module, configured to plan a target trajectory by using the trajectory planning target model, the target trajectory being used to indicate the vehicle to move from a current position to the target parking space; a parking module, configured to control the vehicle to park according to a parking mode corresponding to the target parking scenario and the target trajectory.

11. A computer device, comprising: An apparatus includes one or more processors and memory storing computer-readable instructions that, when executed by the one or more processors, perform the steps of the automatic parking method of any one of claims 1-9.

12. A storage medium, characterized by A storage medium stores computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the automatic parking method of any one of claims 1-9.

Images