Valet parking mapping method and system based on vehicle self-learning and medium

Abstract

The invention relates to a valet parking mapping method and system based on vehicle self-learning and a medium, and the method comprises the steps: map expansion: when a vehicle carries out the map construction of a certain parking lot again, a laser radar and a camera collect environment data again, transmit the environment data to an HAD, after the HAD completes the map construction, upload the map to a cloud end, and send the map to the cloud end; and the cloud side judges whether the two maps have a crossed and overlapped part or not, if the two maps have the crossed and overlapped part and meet the merging requirement, the two maps are merged by adopting a weighted average algorithm based on path finder optimization, and if the two maps do not meet the condition, the two maps are not merged, and the cloud side feeds back a map merging result to the vehicle side. According to the method, the valet parking map is constructed by using the sensor of the vehicle, field end facility construction is not needed, popularization is facilitated, the map learned by the vehicle can be continuously expanded, and the convenience of function use and the driving experience feeling of a user are improved.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to a valet parking mapping method, system, and medium based on vehicle self-learning. Background Technology

[0002] With the continuous development of intelligent connected vehicle technology, valet parking technology, as an important application scenario in the field of autonomous driving, aims to solve the problems of time-consuming and labor-intensive traditional parking processes, difficulties in parking, and high costs of manual valet parking. The primary prerequisite for achieving autonomous valet parking is that the vehicle must obtain map information of its location (or route) to autonomously plan its driving path. Currently, there are two main methods for valet parking map construction in the industry. One method involves the vehicle learning and memorizing parking routes, storing these routes on the vehicle for future use. The second method is a vehicle-parking interaction solution, where sensing devices are deployed within the parking area, and the parking lot sends a complete map of the parking lot to the vehicle via the cloud. Upon receiving the map, the vehicle plans its path and performs the parking maneuver.

[0003] In the prior art, Chinese patent application (application number: 202110975934.7, publication number: CN113724323B) discloses a map construction method, apparatus, and device. The method includes: collecting and saving backtracking data along a backtracking path as the teaching vehicle travels along the backtracking path to the starting point of the task area; activating the map data filling function when the teaching vehicle reaches the starting point of the task area, filling the map data with the backtracking data; collecting task data along the task path from the starting point to the ending point of the task area, and filling the map data with the task data; however, this solution mainly relies on sending parking lot maps from the parking lot terminal to the vehicle terminal for autonomous valet parking, requiring advance data collection and construction of the parking lot area map, and intelligent transformation of the parking lot terminal, including the installation of sensing equipment, resulting in high infrastructure costs and hindering widespread implementation; stopping the map data filling function when the teaching vehicle reaches the ending point of the task area, and constructing a target map based on all the map data.

[0004] In the prior art, Chinese patent application (application number: 202411801582.3, publication number: CN119580526A) discloses an intelligent connected vehicle cloud-cooperative automatic valet parking system and control method. The system includes: a cloud system for scheduling and management; a vehicle-side system including a data fusion processing unit and a planning control unit, wherein the data fusion processing unit is used to fuse and process various data acquired by the vehicle and provide them to the planning control unit; the planning control unit plans parking paths and controls the vehicle to perform parking actions based on the data obtained from the fusion processing; and a field-side system including a field-side visual perception unit, a lidar, and an MEC unit, wherein the field-side visual perception unit and lidar and other sensors are used to perceive the surrounding environment of the vehicle, and the MEC unit is used to process real-time data. The data perceived by the field-side system is processed and analyzed in real time by the MEC to generate decision results. However, this solution cannot be expanded on existing memory maps, the map coverage is limited, and it lacks semantic information about parking spaces on both sides of the route. It can only perform fixed-route parking and driving on a "point-to-point" basis, and cannot select other parking spaces on the map and plan its own route. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the present invention provides a valet parking mapping method, system and medium based on vehicle self-learning. It not only uses the vehicle's own sensors to build a valet parking map without the need for site infrastructure construction, which is conducive to promotion, but also can continuously expand the map learned by the vehicle itself, thereby improving the convenience of use and the user's driving experience.

[0006] To achieve the above and other related objectives, the present invention provides the following technical solution: A valet parking mapping method based on vehicle self-learning, the method comprising: U1. Initial Map Building: When a vehicle first enters a parking lot for map building, LiDAR and cameras collect environmental data and transmit it to HAD. After HAD completes map building, it uploads the map to the cloud, where the cloud manages the uploaded map. U2. Map Expansion: When the vehicle rebuilds a map of a parking lot again, the LiDAR and camera collect environmental data again and transmit it to HAD. After the HAD completes the map building, it uploads the map to the cloud. The cloud determines whether there is any overlap between the two maps. If there is and the merging requirements are met, the two maps are merged using a weighted average algorithm based on Pathfinder optimization. If the conditions are not met, the maps are not merged, and the cloud feeds back the map merging result to the vehicle. U3. Vehicle-side parking in or out: When using the valet parking function, after the user enables the function on the vehicle-side HMI, the HAD initiates a map data request to the cloud, and the cloud returns the corresponding map data to the HAD to support the vehicle-side in completing parking in and out operations.

[0007] Furthermore, the method also includes: U4. Parking in or out via APP: When using the parking function of the mobile APP, after the user activates the function in the APP, the APP requests map data from the cloud. The cloud synchronously returns the map data to HAD and the APP. The APP completes the map loading and display, and the vehicle plans the route according to the map, supporting the APP to complete parking in and out operations.

[0008] Furthermore, in step U2, the two maps include data information of the first parking lot construction map and data information of the second parking lot construction map.

[0009] Furthermore, the method of merging the two maps using a weighted average algorithm based on Pathfinder optimization includes: U21. Based on the data information of the first parking lot construction map and the data information of the second parking lot construction map, normalization processing is performed to obtain the normalized data information of the first parking lot construction map and the data information of the second parking lot construction map; U22. Based on the data information from the first and second parking lot map constructions after normalization, a weighted average fusion function W is constructed. , Where, x 1i To construct the first parking lot map using coordinate point data after normalization, x 2j The coordinate point data information for constructing the map of the second parking lot after normalization, α1, α2, α3, ..., α i ,...,ɑ M and β1, β2, β3, ..., β j , ..., β N These are the weighting coefficients, where M and N are positive integers; U23. Based on the weighted average fusion function W, the Pathfinder optimization algorithm is used to optimize the weight coefficients α1, α2, α3, ..., α. i ,...,ɑ M and β1, β2, β3, ..., β j , ..., β N The optimized weighted average fusion function W is obtained through optimization, and the two maps are then merged.

[0010] Furthermore, in step U23, the Pathfinder optimization algorithm is used to optimize the weight coefficients α1, α2, α3, ..., α. i ,...,ɑ M and β1, β2, β3, ..., β j , ..., β N Optimization includes: U231. Based on the weighting coefficients α1, α2, α3, ..., α i ,...,ɑ M and β1, β2, β3, ..., β j , ..., β N The population is initialized, and the population parameters and the maximum number of iterations K are determined. max This yields the data information of the initialized population. U232. Based on the data information of the initialized population, calculate the fitness and determine the searchers and followers; U233. According to function g i K+1 , , Where K is the current iteration number, r1 is the pathfinder's step size factor, and μ1 is any constant parameter between -1 and 1. The searcher's position is updated according to the function h. j K+1 , , Where r2 and r3 are the step size factors for moving with other followers and pathfinders, and ρ is any constant parameter between 0 and 1, which updates the position of the followers; U234. Calculate the fitness value, update the global optimum, and determine whether the maximum number of iterations K has been reached. max If the optimal value is achieved, output the optimal value; otherwise, repeat steps U232-U234.

[0011] Furthermore, the step size factor r1 of the pathfinder movement is any constant parameter between 0 and 1, and the constraint conditions for the step size factors r2 and r3 of the pathfinder movement with other followers and pathfinders are any constant parameters between 0 and 1.

[0012] Furthermore, the maximum number of iterations K max Greater than zero.

[0013] To achieve the above and other related objectives, the present invention also provides a valet parking mapping system based on vehicle self-learning, used to implement the aforementioned valet parking mapping method based on vehicle group learning. The system includes sensors, a HAD (Hybrid Access Device), a cloud platform, and a mobile app. The sensor is used by the vehicle to collect driving route and environmental information through lidar and camera sensors and send it to HAD for information processing. The HAD is connected to the sensor and is used by the autonomous driving system to build and generate a structured map based on the environmental information collected by the vehicle's sensors. The cloud is connected to the HAD and is used to store and manage the maps generated by the vehicle-side HAD. The expansion and merging of the maps are carried out in the cloud. The app is connected to the cloud and is used by users to activate the valet parking function using a mobile app.

[0014] Furthermore, the sensor also includes a millimeter-wave radar sensor for acquiring the distance between obstacles and the vehicle.

[0015] To achieve the above and other related objectives, the present invention also provides a computer-readable storage medium storing a computer program programmed or configured to perform the described vehicle self-learning-based valet parking mapping method.

[0016] The present invention has the following positive effects: This invention determines whether two maps overlap via the cloud. If they do and meet the merging requirements, the two maps are merged using a weighted average algorithm optimized by Pathfinder. If the conditions are not met, they are not merged. The cloud then feeds the map merging result back to the vehicle. This not only utilizes the vehicle's own sensors to build valet parking maps without requiring site infrastructure construction, making it easy to promote, but also allows for continuous expansion of the maps learned by the vehicle, improving the convenience of use and the user's driving experience. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the system framework of the present invention; Figure 2 This is a schematic diagram of the method flow of the present invention; Figure 3 This is a flowchart illustrating the weighted average algorithm based on Pathfinder Optimization of the present invention. Detailed Implementation

[0018] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0019] Example 1: As Figure 2 As shown, a valet parking mapping method based on vehicle self-learning is described, the method comprising: U1. Initial Map Building: When a vehicle first enters a parking lot for map building, LiDAR and cameras collect environmental data and transmit it to HAD. After HAD completes map building, it uploads the map to the cloud, where the cloud manages the uploaded map. U2. Map Expansion: When the vehicle rebuilds a map of a parking lot again, the LiDAR and camera collect environmental data again and transmit it to HAD. After the HAD completes the map building, it uploads the map to the cloud. The cloud determines whether there is any overlap between the two maps. If there is and the merging requirements are met, the two maps are merged using a weighted average algorithm based on Pathfinder optimization. If the conditions are not met, the maps are not merged, and the cloud feeds back the map merging result to the vehicle. U3. Vehicle-side parking in or out: When using the valet parking function, after the user enables the function on the vehicle-side HMI, the HAD initiates a map data request to the cloud, and the cloud returns the corresponding map data to the HAD to support the vehicle-side in completing parking in and out operations.

[0020] In this embodiment, the method further includes: U4. Parking in or out via APP: When using the parking function of the mobile APP, after the user activates the function in the APP, the APP requests map data from the cloud. The cloud synchronously returns the map data to HAD and the APP. The APP completes the map loading and display, and the vehicle plans the route according to the map, supporting the APP to complete parking in and out operations.

[0021] In this embodiment, in step U2, the two maps include data information of the first parking lot construction map and data information of the second parking lot construction map.

[0022] In this embodiment, as Figure 3 As shown, the process of merging the two maps using a weighted average algorithm based on Pathfinder optimization includes: U21. Based on the data information of the first parking lot construction map and the data information of the second parking lot construction map, normalization processing is performed to obtain the normalized data information of the first parking lot construction map and the data information of the second parking lot construction map; U22. Based on the data information from the first and second parking lot map constructions after normalization, a weighted average fusion function W is constructed. , Where, x 1i To construct the first parking lot map using coordinate point data after normalization, x 2jThe coordinate point data information for constructing the map of the second parking lot after normalization, α1, α2, α3, ..., α i ,...,ɑ M and β1, β2, β3, ..., β j , ..., β N These are the weighting coefficients, where M and N are positive integers; U23. Based on the weighted average fusion function W, the Pathfinder optimization algorithm is used to optimize the weight coefficients α1, α2, α3, ..., α. i ,...,ɑ M and β1, β2, β3, ..., β j , ..., β N The optimized weighted average fusion function W is obtained through optimization, and the two maps are then merged.

[0023] In this embodiment, in step U23, the Pathfinder optimization algorithm is used to optimize the weight coefficients α1, α2, α3, ..., α. i ,...,ɑ M and β1, β2, β3, ..., β j , ..., β N Optimization includes: U231. Based on the weighting coefficients α1, α2, α3, ..., α i ,...,ɑ M and β1, β2, β3, ..., β j , ..., β N The population is initialized, and the population parameters and the maximum number of iterations K are determined. max This yields the data information of the initialized population. U232. Based on the data information of the initialized population, calculate the fitness and determine the searchers and followers; U233. According to function g i K+1 , , Where K is the current iteration number, r1 is the pathfinder's step size factor, and μ1 is any constant parameter between -1 and 1. The searcher's position is updated according to the function h. j K+1 , , Where r2 and r3 are the step size factors for moving with other followers and pathfinders, and ρ is any constant parameter between 0 and 1, which updates the position of the followers; U234. Calculate the fitness value, update the global optimum, and determine whether the maximum number of iterations K has been reached.max If the optimal value is achieved, output the optimal value; otherwise, repeat steps U232-U234.

[0024] In this embodiment, the step size factor r1 of the pathfinder movement is any constant parameter between 0 and 1, and the constraint conditions for the step size factors r2 and r3 of the pathfinder movement with other followers and pathfinders are any constant parameters between 0 and 1.

[0025] In this embodiment, the maximum number of iterations K max Greater than zero.

[0026] Example 2: Based on the vehicle self-learning-based valet parking mapping method in Example 1, the present invention will be further explained and described below.

[0027] like Figure 2 As shown, a valet parking mapping method based on vehicle self-learning is described, the method comprising: U1. Initial Map Building: When a vehicle first enters a parking lot for map building, LiDAR and cameras collect environmental data and transmit it to HAD. After HAD completes map building, it uploads the map to the cloud. The cloud manages the uploaded map. U2. Map Expansion: When the vehicle rebuilds a map of a parking lot again, the LiDAR and camera collect environmental data again and transmit it to HAD. After the HAD completes the map building, it uploads the map to the cloud. The cloud determines whether there is any overlap between the two maps. If there is and the merging requirements are met, the two maps are merged using a weighted average algorithm based on Pathfinder optimization. If the conditions are not met, the maps are not merged, and the cloud feeds back the map merging result to the vehicle. U3. Vehicle-side parking in or out: When using the valet parking function, after the user enables the function on the vehicle-side HMI, the HAD initiates a map data request to the cloud, and the cloud returns the corresponding map data to the HAD to support the vehicle-side in completing parking in and out operations.

[0028] like Figure 1 As shown, this invention provides a valet parking mapping system based on vehicle self-learning, used to implement the aforementioned valet parking mapping method based on vehicle group learning. The system includes sensors, HAD (Hardware-Aided Utility), cloud computing, and a mobile app. The sensor is used by the vehicle to collect driving route and environmental information through lidar and camera sensors and send it to HAD for information processing. The HAD is connected to the sensor and is used by the autonomous driving system to build and generate a structured map based on the environmental information collected by the vehicle's sensors. The cloud is connected to the HAD and is used to store and manage the maps generated by the vehicle-side HAD. The expansion and merging of the maps are carried out in the cloud. The app is connected to the cloud and is used by users to activate the valet parking function using a mobile app.

[0029] In this embodiment, the sensor also includes a millimeter-wave radar sensor for acquiring the distance between the obstacle and the vehicle.

[0030] In this embodiment, the present invention provides a computer-readable storage medium storing a computer program programmed or configured to perform the described vehicle self-learning-based valet parking mapping method.

[0031] Any references to memory, storage, database, or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0032] In summary, this invention not only utilizes the vehicle's own sensors to construct valet parking maps without requiring site infrastructure construction, thus facilitating widespread adoption, but also allows for continuous expansion of the maps learned by the vehicle itself, enhancing the convenience of use and the user's driving experience.

[0033] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for vehicle self-learning-based valet parking mapping, characterized in that, The method comprises: U1. Initial map construction: when a vehicle performs map construction in a parking lot for the first time, laser radar and camera collect environment data, which is transmitted to HAD, and HAD completes map construction and uploads the map to the cloud, and the cloud manages the uploaded map; U2. Map expansion: when the vehicle performs map construction in the parking lot again, laser radar and camera collect environment data again, which is transmitted to HAD, and HAD completes map construction and uploads the map to the cloud, and the cloud judges whether the two maps have overlapping parts, if there are overlapping parts and the merging requirements are met, the two maps are merged by using a weighted average algorithm based on explorer optimization, if the conditions are not met, the two maps are not merged, and the cloud feeds back the map merging result to the vehicle end; U3. Vehicle end parking in or parking out: when the valet parking function is used, the user starts the function on the vehicle end HMI, HAD initiates a map data request to the cloud, the cloud returns corresponding map data to HAD, and the vehicle end completes parking in and parking out operations.

2. The method of claim 1, wherein the method is based on self-learning of a vehicle. The method further comprises: U4. APP parking in or parking out: when the APP parking function is used, the user starts the function on the APP, the APP requests map data from the cloud, the cloud synchronously returns the map data to HAD and the APP, the APP completes map loading and display, the vehicle plans a path according to the map route, and the APP end completes parking in and parking out operations. 3.The method of claim 1, wherein, In step U2, the two maps comprise data information of the first parking lot construction map and data information of the second parking lot construction map.

4. The method of claim 3, wherein the method further comprises: The merging of the two maps by using the weighted average algorithm based on explorer optimization comprises: U21. Based on the data information of the first parking lot construction map and the data information of the second parking lot construction map, normalization processing is performed to obtain normalized data information of the first parking lot construction map and normalized data information of the second parking lot construction map; U22. Based on the normalized data information of the first parking lot construction map and the normalized data information of the second parking lot construction map, a weighted average fusion function W is constructed, ， wherein x 1i is the coordinate point data information of the first parking lot map construction after normalization processing, x 2j is the coordinate point data information of the second parking lot map construction after normalization processing, a1, a2, a3,..., a i ,..., a M and b1, b2, b3,..., b j ,..., b N are weight coefficients, and M and N are positive integers; U23. Based on the weighted average fusion function W, the weight coefficients a1, a2, a3,..., a i ,...,a M and β1, β2, β3,..., β j ,...,β N are optimized by using the pathfinder optimization algorithm to obtain the optimized weighted average fusion function W, and the two maps are merged.

5. The method of claim 4, wherein the method further comprises: In step U23, the employing the scout optimization algorithm to optimize the weight coefficients a1, a2, a3,..., a i ,...,a M and β1, β2, β3,..., β j ,...,β N includes: U231. based on the weight coefficients a1, a2, a3,..., a i ,..., a M and b1, b2, b3,..., b j ,..., b N , initialize the population, determine the population parameters and the maximum number of iterations K max , obtain the data information of the initialized population; U232. Based on the data information of the initialized population, the fitness is calculated to determine the explorer and the follower; U233. According to the function g i K+1 , , where K is the current iteration number, r1 is a step factor of the pathfinder movement, μ1 is an arbitrary constant parameter between -1 and 1, the position of the searchers is updated according to the function h j K+1 , ， wherein r2 and r3 are step length factors of movement of other followers and explorers, and p is an arbitrary constant parameter between 0 and 1, and the position of the follower is updated; U234. Calculate fitness value, and update global optimal value, judge whether to reach the maximum iteration number K max If reached, output the optimal value, otherwise repeat steps U232-U234.

6. The vehicle self-learning-based valet parking mapping method according to claim 5, characterized in that: The step length factor r1 of movement of the explorer ranges from 0 to 1, and the constraint conditions of the step length factors r2 and r3 of movement of other followers and explorers are any constant parameters between 0 and 1.

7. The method of claim 5, wherein the method further comprises: the maximum number of iterations K max greater than zero.

8. A valet parking mapping system based on vehicle self-learning, characterized in that, The system for implementing the vehicle group learning-based valet parking map construction method of any one of claims 1-7 comprises sensors, HAD, a cloud and an APP, The sensors are configured to collect driving routes and environment information by laser radar and camera sensors and transmit the information to HAD for information processing; The HAD is connected with the sensors and is configured to construct and generate a structured map according to the environment information collected by the vehicle sensors. The cloud is connected with the HAD and is used for storing and managing the map generated by the HAD, and expansion and merging of the map are performed in the cloud. The APP is in communication connection with the cloud and is used for starting the valet parking function by the user by using the mobile phone APP. 9.The valet parking mapping system based on vehicle self-learning according to claim 8, characterized in that, The sensor further comprises a millimeter wave radar sensor for collecting the distance between the obstacle and the vehicle.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program programmed or configured to perform the vehicle self-learning-based valet parking mapping method according to any one of claims 1-7.

Images