Autonomous parking system and method based on visual simultaneous localization and mapping

Abstract

The present application relates to the technical field of autonomous parking, in particular to an autonomous parking system and method based on visual simultaneous localization and mapping, wherein a car owner first starts one-key parking, the system performs self-checking and responds to the parking command of the car owner, then inputs an RGB-D image stream into a visual simultaneous localization and mapping module to realize positioning and scene perception of the vehicle, then adopts a target detection algorithm to detect idle parking spaces, scores the idle parking spaces, and selects idle parking spaces with scores higher than a set threshold as target parking spaces, finally determines an optimal path by using a path planning algorithm, guides the vehicle to the target parking space and performs parking operation, and the vehicle is turned off, and the one-key car calling function can also be realized, which can not only find the best path, but also consider the existence of static and dynamic obstacles, and improve the safety of driving.

Description

Technical Field

[0001] This invention relates to the field of autonomous parking technology, and in particular to an autonomous parking system and method based on visual synchronous localization and mapping. Background Technology

[0002] To improve parking efficiency, reduce parking accidents, and provide drivers with a safer and more convenient driving experience, autonomous parking systems have emerged. However, autonomous parking systems face numerous challenges, including vehicle localization, scene perception, path planning, and vehicle control. Among these, vehicle localization and scene perception are two key technical issues. Traditional autonomous parking systems typically rely on the integrated use of data from multiple sensors, such as fisheye cameras and LiDAR, to solve the technical difficulties of vehicle localization and scene perception. Existing technologies, through sensor fusion algorithms and point cloud registration, enable these systems to provide vehicle localization information and perceive and identify various obstacles and available parking spaces in the parking lot environment. Based on this, the autonomous parking system performs path planning so that the vehicle can effectively drive to the target parking space.

[0003] Traditional solutions based on LiDAR or GDS (Global Navigation Satellite System) suffer from limitations in positioning accuracy, high hardware costs, and unstable signals, and struggle to meet real-time requirements and adapt to rapidly changing scenarios. Specifically, in vehicle localization and scene perception, multi-source fusion and point cloud registration using sensors such as GDS and LiDAR require significant computational resources and time to process and analyze data, leading to reduced real-time performance and increased response time latency. Furthermore, the low signal accuracy of GDS systems can easily result in erroneous vehicle localization and scene perception. Meanwhile, their poor environmental adaptability cannot be ignored. In complex parking lot environments, numerous dynamic obstacles and scene changes cause insufficient accuracy in vehicle localization and environmental perception of traditional autonomous parking systems. Moreover, the path planning algorithms used in existing autonomous parking systems (such as A*, D*, and fast exploratory random trees) are only suitable for static or small-scale dynamic scenarios, and are prone to errors or even accidents in large-scale dynamic scenarios like parking lots. Summary of the Invention

[0004] To address the issues of poor real-time performance and low accuracy in vehicle positioning and environmental perception, this invention proposes an autonomous parking system and method based on simultaneous localization and mapping (SLAM). Based on scene information, the system achieves autonomous parking and vehicle retrieval through SLAM, target detection technology, and path planning strategies.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] The first invention provides an autonomous parking system based on simultaneous localization and mapping (SLAM), comprising:

[0007] The one-click parking module includes a one-click parking button, which allows the car owner to initiate the autonomous parking process by pressing the button.

[0008] The synchronous localization and mapping module uses a dense visual synchronous localization and mapping model based on neural implicit scalable coding to locate the vehicle and build a scene map in real time to accurately reflect environmental information.

[0009] The vacant parking space detection module uses a target detection model to detect parking spaces and identify vacant parking spaces.

[0010] The parking space evaluation and optimal parking space selection module selects the target parking space as the optimal parking space based on the parking space evaluation function.

[0011] The path planning module uses path planning algorithms to determine the optimal driving path and autonomously avoid obstacles.

[0012] The one-click car-calling module includes a one-click car-calling button, which allows car owners to initiate the autonomous car-calling process by selecting the one-click car-calling button.

[0013] The self-test module for autonomous vehicles determines whether an autonomous vehicle is usable and sends the results back to the owner.

[0014] The autonomous vehicle positioning and path planning module receives positioning information sent by the vehicle owner, determines the owner's current location, and uses the same method as the synchronous positioning and mapping module to locate the vehicle and build a scene map. Then, based on the owner's location and the scene map, it performs path planning to determine the optimal path.

[0015] Secondly, the present invention provides an autonomous parking method based on simultaneous localization and mapping, specifically including the following steps:

[0016] S1. The car owner can start the automatic parking process through the one-click parking module;

[0017] S2. After receiving the one-click parking command, the autonomous parking system inputs the RGB-D image stream acquired by the RGB-D camera into the synchronous localization and mapping module. The synchronous localization and mapping module uses a dense visual synchronous localization and mapping model based on neural implicit scalable coding to locate the vehicle, build a scene map in real time, and output the camera pose and the scene representation in the form of a learned hierarchical feature grid.

[0018] S3. After obtaining the hierarchical feature grid, the target detection model is used to detect parking spaces to determine whether there are any available parking spaces. If there are no available parking spaces, the parking space detection is repeated; if there are available parking spaces, the parking space is evaluated.

[0019] S4. Use a parking space evaluation function to score vacant parking spaces and select vacant parking spaces with scores higher than a set threshold as target parking spaces.

[0020] S5. After determining the target parking space, a path planning algorithm is used to determine the optimal path and autonomously avoid obstacles to guide the vehicle to the target parking space.

[0021] S6. The vehicle enters the parking space according to the planned optimal path, and after parking, the vehicle is turned off, completing the autonomous parking process.

[0022] As a further technical solution of the present invention, the layered feature mesh in step S2 is composed of a coarse geometric mesh. Intermediate geometric grid Fine geometric mesh and color grid It consists of four parts, with side lengths of 200cm, 32cm, 16cm, and 16cm respectively for the four feature grids. The corresponding decoders for the four feature grids are as follows: , , and With a fixed pre-trained geometric decoder, only the parameters of the feature mesh are optimized. and color decoder The parameters are used to generate corresponding depth and RGB images from the hierarchical feature mesh through the hierarchical encoder and hierarchical renderer. The optimization goal is to minimize the image reconstruction loss, that is, to maximize the similarity between the generated image and the real image. The hierarchical feature mesh and camera pose corresponding to the minimum image reconstruction loss are the final outputs of the dense visual synchronous localization and mapping model.

[0023] As a further technical solution of the present invention, the process of generating corresponding depth images and RGB images from the layered feature mesh via a layered encoder and a layered renderer in step S2 is as follows:

[0024] (1) From hierarchical feature mesh to hierarchical decoder, for any given point , This represents the coarse geometric features of the point within a coarse geometric mesh, as determined by the corresponding decoder. The predicted occupancy probability of that point on the rough layer is then obtained. Similarly, we get points. The predicted occupancy probability on the intermediate layer is ,point Predicting occupancy probability at the fine-grained level Indicate, then point The predicted occupancy probabilities from the intermediate and fine layers are superimposed to obtain... The features in the color grid are decoded to obtain the points. Color prediction ;

[0025] (2) From the layered decoder to the layered renderer, the layered renderer processes the coarse geometric information separately. and fine geometric information Perform depth rendering and process color information. Perform color rendering, so that Indicates the origin of the given camera The first ray on the ray There are 1 sampling points, among which Corresponding to the point along this ray The depth value, That is, uniform sampling of each ray. One point, point The coarse occupancy probability is output by the layered decoder. Fine occupancy probability and color prediction For each ray, coarse depth rendering, fine depth rendering, and color rendering are respectively used... , and It means that, among them, , , ; , .

[0026] As a further technical solution of the present invention, the process of minimizing the image reconstruction loss in step S2 is as follows:

[0027] First, create a global keyframe list, and then add new keyframes gradually based on information gain. Next, sample from the current frame and keyframes. 1 pixel and Representing its true depth and color information respectively, and finally calculating and minimizing this... The reconstruction loss per pixel includes geometric loss. and light loss Among them, geometric loss Due to rough geometry loss and fine geometric loss It consists of two parts. , , ;

[0028] Then, a real-time multi-stage optimization strategy is adopted to minimize the reconstruction loss, first minimizing the coarse geometric loss. To optimize the intermediate layer geometry mesh, and then by minimizing We work together to optimize the intermediate and fine-grained geometries, and finally minimize the total loss. To jointly optimize the geometric mesh, color decoder, and camera extrinsic parameters of K selected keyframes at all levels, where , It is used to balance the loss of light intensity. The hyperparameters of the camera, the camera extrinsic parameters, are used to describe the camera's attitude and position in the world coordinate system, specifically the camera's rotation matrix and translation vector.

[0029] As a further technical solution of the present invention, the target detection model in step S3 is a trained YoLo model or a Mask R-CNN model.

[0030] As a further technical solution of the present invention, the parking space evaluation function in step S4 is: ,in, , and These represent the size, availability, and distance of the parking space, respectively. , and This is a hyperparameter.

[0031] As a further technical solution of the present invention, the path planning algorithm in step S5 adopts the RND3QN algorithm.

[0032] Thirdly, the present invention provides an autonomous vehicle hailing method based on simultaneous localization and mapping, comprising the following steps:

[0033] First, the car owner initiates autonomous vehicle shunting via the one-click vehicle shunting module. The autonomous driving vehicle automatically determines whether the vehicle is available. If the vehicle is unavailable, it notifies the owner that the vehicle is in an unavailable state. If the vehicle is available, the system receives the location information sent by the owner. During the shunting process, the system uses the autonomous driving vehicle positioning and route planning module to locate the vehicle and perceive the environment, and after determining the optimal route, it drives to the location specified by the owner. Then, it automatically determines whether it has reached the target location. If it has not reached the target location, it continues driving. If it has reached the target location, the vehicle automatically stops and turns off the engine, completing the one-click vehicle shunting process.

[0034] Compared with the prior art, the beneficial effects of the present invention are:

[0035] This invention provides an autonomous parking system and method based on simultaneous localization and mapping (SMR) to solve the problems of autonomous vehicle parking and vehicle retrieval, thereby improving the real-time performance and accuracy of vehicle localization and scene perception. This is mainly reflected in three aspects:

[0036] 1. High real-time performance: By adopting dense vision-based synchronous localization and mapping technology, vehicle localization and scene mapping can be achieved faster, thereby significantly improving the real-time performance of the system.

[0037] 2. High precision: Based on dense vision synchronous positioning and mapping technology, it can accurately capture the geometric structure and appearance details of the scene, realize precise perception of the environment, and significantly improve the accuracy of vehicle positioning and scene perception.

[0038] 3. Robust path planning algorithm: The autonomous parking system designed in this invention can not only find the optimal path, but also take into account the existence of static and dynamic obstacles. Such intelligent planning enables the vehicle to avoid obstacles and improve driving safety. Attached Figure Description

[0039] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain this disclosure and do not constitute an undue limitation of the invention.

[0040] Figure 1 The diagram shows the structure of the autonomous parking system based on synchronous positioning and mapping provided by this invention.

[0041] Figure 2 This is a schematic diagram illustrating the process of autonomous parking in the autonomous parking system based on synchronous positioning and mapping provided by the present invention.

[0042] Figure 3 This is a schematic diagram of the autonomous parking system based on synchronous positioning and mapping provided by the present invention during autonomous vehicle retrieval.

[0043] Figure 4 This is a schematic diagram of the synchronous positioning and mapping process provided by the present invention. Detailed Implementation

[0044] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0045] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0046] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. Furthermore, it should be understood that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0047] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0048] Example 1:

[0049] like Figure 1 As shown, an autonomous parking system based on simultaneous localization and mapping (SLAM) in an embodiment of the present invention includes:

[0050] The one-click parking module includes a one-click parking button, which allows the car owner to initiate the autonomous parking process by pressing the button.

[0051] The synchronous localization and mapping module uses a dense visual synchronous localization and mapping model based on neural implicit scalable coding to locate the vehicle and build a scene map in real time to accurately reflect environmental information.

[0052] The vacant parking space detection module uses a target detection model to detect parking spaces and identify vacant parking spaces.

[0053] The parking space evaluation and optimal parking space selection module selects the target parking space as the optimal parking space based on the parking space evaluation function.

[0054] The path planning module uses path planning algorithms to determine the optimal driving path and autonomously avoid obstacles.

[0055] The one-click car-calling module includes a one-click car-calling button, which allows car owners to initiate the autonomous car-calling process by selecting the one-click car-calling button.

[0056] The self-test module for autonomous vehicles determines whether an autonomous vehicle is usable and sends the results back to the owner.

[0057] The autonomous vehicle positioning and path planning module receives positioning information sent by the vehicle owner, determines the owner's current location, and uses the same method as the synchronous positioning and mapping module to locate the vehicle and build a scene map. Then, based on the owner's location and the scene map, it performs path planning to determine the optimal path. In this process, it ensures the real-time performance and robustness of the system, quickly responds to user operation commands and environmental changes, and adapts to dynamic large-scale scenes and noisy environments.

[0058] Example 2:

[0059] like Figure 2 and Figure 3 As shown, an autonomous parking process based on simultaneous localization and mapping (SMR) in an embodiment of the present invention includes the following steps:

[0060] Specifically, the following steps are included:

[0061] S1. The car owner can start the automatic parking process through the one-click parking module;

[0062] S2. After receiving the one-click parking command, the autonomous parking system inputs the RGB-D image stream acquired by the RGB-D camera into the synchronous localization and mapping module. The synchronous localization and mapping module uses a dense visual synchronous localization and mapping model based on neural implicit scalable coding to locate the vehicle, build a scene map in real time, and output the camera pose and the scene representation in the form of a learned hierarchical feature grid. The hierarchical feature grid consists of a coarse geometric grid. Intermediate geometric grid Fine geometric mesh and color grid It consists of four parts, with side lengths of 200cm, 32cm, 16cm, and 16cm respectively for the four feature grids. The corresponding decoders for the four feature grids are as follows: , , and With a fixed pre-trained geometric decoder, only the parameters of the feature mesh are optimized. and color decoder The parameters are used to generate corresponding depth and RGB images from the hierarchical feature mesh via a hierarchical encoder and a hierarchical renderer. The optimization objective is to minimize the image reconstruction loss, i.e., to maximize the similarity between the generated image and the real image. The hierarchical feature mesh and camera pose corresponding to the minimum image reconstruction loss are the final outputs of the dense visual simultaneous localization and mapping model, specifically:

[0063] (1) From hierarchical feature mesh to hierarchical decoder, for any given point , This represents the coarse geometric features of the point within a coarse geometric mesh, as determined by the corresponding decoder. The predicted occupancy probability of that point on the rough layer is then obtained. Similarly, we get points. The predicted occupancy probability on the intermediate layer is ,point Predicting occupancy probability at the fine-grained level Indicate, then point The predicted occupancy probabilities from the intermediate and fine layers are superimposed to obtain... The features in the color grid are decoded to obtain the points. Color prediction ;

[0064] (2) From the layered decoder to the layered renderer, the layered renderer processes the coarse geometric information separately. and fine geometric information Perform depth rendering and process color information. Perform color rendering, so that Indicates the origin of the given camera The first ray on the ray There are 1 sampling points, among which Corresponding to the point along this ray The depth value, That is, uniform sampling of each ray. One point, point The coarse occupancy probability is output by the layered decoder. Fine occupancy probability and color prediction For each ray, coarse depth rendering, fine depth rendering, and color rendering are respectively used... , and It means that, among them, , , ; , ;

[0065] (3) To minimize image reconstruction loss, first establish a global keyframe list, then add new keyframes step by step according to information gain, and then sample from the current frame and keyframes. 1 pixel and Representing its true depth and color information respectively, and finally calculating and minimizing this... The reconstruction loss per pixel includes geometric loss. and light loss Among them, geometric loss Due to rough geometry loss and fine geometric loss It consists of two parts. , , ;

[0066] Then, a real-time multi-stage optimization strategy is adopted to minimize the reconstruction loss, first minimizing the coarse geometric loss. To optimize the intermediate layer geometry mesh, and then by minimizing We work together to optimize the intermediate and fine-grained geometries, and finally minimize the total loss. To jointly optimize the geometric mesh, color decoder, and camera extrinsic parameters of K selected keyframes at all levels, where , It is used to balance the loss of light intensity. The hyperparameters of the camera, the camera extrinsic parameters, are used to describe the camera's attitude and position in the world coordinate system, specifically the camera's rotation matrix and translation vector.

[0067] This multi-stage optimization scheme can converge better because higher resolution appearance (color) and fine layer features can rely on already refined geometry from intermediate layer feature meshes. When the camera moves to previously unobserved areas, the real-time optimized coarse feature meshes provide meaningful information for camera tracking, making it more robust and reliable to sudden frame loss or rapid camera movement.

[0068] S3. The layered feature mesh obtained in step S2 represents the depth and color information of the surrounding environment under the current camera pose. After obtaining the layered feature mesh, the trained object detection model (YoLo, Mask R-CNN, etc.) is used to detect parking spaces in the scene and obtain the accurate location and boundary information of the parking spaces, such as the width and length of the parking spaces and the distance between the parking spaces and the vehicle. If there are no vacant parking spaces, the parking space detection is repeated; if there are vacant parking spaces, the parking space is evaluated.

[0069] S4. After detecting an available parking space, score the available parking spaces using a parking space evaluation function, and select available parking spaces with scores higher than a set threshold as target parking spaces; the parking space evaluation function is: ,in, , and These represent the size, availability, and distance of the parking space, respectively. Generally speaking, larger parking spaces are easier to park, parking spaces that are not occupied by any vehicles or obstacles are more available, and parking spaces that are closest to your vehicle can save driving time and energy. , and Hyperparameters are manually set.

[0070] S5. After determining the target parking space, a path planning algorithm is used to determine the optimal path and autonomously avoid obstacles, guiding the vehicle to the target parking space. In this embodiment, the RND3QN path planning algorithm is used to plan the optimal parking route. First, the data is fed into the pre-trained RND3QN model. Then, the RND3QN model selects the action with the highest Q value as the next action of the vehicle (going straight, braking, etc.). While moving towards the target position, it can avoid static or dynamic obstacles in real time. The RND3QN model is a path planning model based on reinforcement learning, which overcomes the shortcomings of traditional path planning algorithms such as Dijkstra's algorithm and A* algorithm, which are only applicable to static scenes or small dynamic scenes.

[0071] S6. The vehicle enters the parking space according to the planned optimal path, and after parking, the vehicle is turned off, completing the autonomous parking process.

[0072] Example 3:

[0073] like Figure 4 As shown, an autonomous vehicle hailing method based on synchronous localization and mapping in an embodiment of the present invention includes the following steps:

[0074] First, the car owner initiates autonomous vehicle shunting via the one-click vehicle shunting module. The autonomous driving vehicle automatically determines whether the vehicle is available. If the vehicle is unavailable, for example, due to insufficient power, it will notify the owner that the vehicle is unavailable. If the vehicle is available, the system receives the location information sent by the owner. During the shunting process, the system uses the autonomous driving vehicle positioning and route planning module to locate the vehicle and perceive the environment, and after determining the optimal route, it drives to the location specified by the owner. Then, it automatically determines whether it has reached the target location. If it has not reached the target location, it continues driving. If it has reached the target location, the vehicle automatically stops and turns off the engine, completing the one-click vehicle shunting process.

[0075] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0076] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. An autonomous parking method based on simultaneous localization and mapping, characterized in that, Specifically, the following steps are included: S1. The car owner can start the automatic parking process through the one-click parking module; S2. After receiving the one-click parking command, the autonomous parking system inputs the RGB-D image stream acquired by the RGB-D camera into the synchronous localization and mapping module. The synchronous localization and mapping module uses a dense visual synchronous localization and mapping model based on neural implicit scalable coding to locate the vehicle, build a scene map in real time, and output the camera pose and the scene representation in the form of a learned hierarchical feature grid; the hierarchical feature grid consists of a coarse geometric grid. Intermediate geometric grid Fine geometric mesh and color grid It consists of four parts, with side lengths of 200cm, 32cm, 16cm, and 16cm respectively for the four feature grids. The corresponding decoders for the four feature grids are as follows: , , and With a fixed pre-trained geometric decoder, only the parameters of the feature mesh are optimized. and color decoder The parameters are used to generate corresponding depth and RGB images through a hierarchical encoder and a hierarchical renderer. The optimization goal is to minimize the image reconstruction loss, that is, to maximize the similarity between the generated image and the real image. The hierarchical feature grid and camera pose corresponding to the minimum image reconstruction loss are the final output of the dense visual synchronous localization and mapping model. S3. After obtaining the hierarchical feature grid, the target detection model is used to detect parking spaces to determine whether there are any available parking spaces. If there are no available parking spaces, the parking space detection is repeated. If there are available parking spaces, an assessment of the parking spaces will be conducted; S4. Use a parking space evaluation function to score vacant parking spaces and select vacant parking spaces with scores higher than a set threshold as target parking spaces. S5. After determining the target parking space, a path planning algorithm is used to determine the optimal path and autonomously avoid obstacles to guide the vehicle to the target parking space. S6. The vehicle enters the parking space according to the planned optimal path, and after parking, the vehicle is turned off, completing the autonomous parking process.

2. The autonomous parking method based on simultaneous localization and mapping according to claim 1, characterized in that, In step S2, the process of generating the corresponding depth image and RGB image from the layered feature mesh via the layered encoder and layered renderer is as follows: (1) From hierarchical feature mesh to hierarchical decoder, for any given point , This represents the coarse geometric features of the point within a coarse geometric mesh, as determined by the corresponding decoder. The predicted occupancy probability of that point on the rough layer is then obtained. Similarly, we get points. The predicted occupancy probability on the intermediate layer is ,point Predicting occupancy probability at the fine-grained level Indicate, then point The predicted occupancy probabilities from the intermediate and fine layers are superimposed to obtain... The features in the color grid are decoded to obtain the points. Color prediction ; (2) From the layered decoder to the layered renderer, the layered renderer processes the coarse geometric information separately. and fine geometric information Perform depth rendering and process color information. Perform color rendering, so that Indicates the origin of the given camera The first ray on the ray There are 1 sampling points, among which Corresponding to the point along this ray The depth value, That is, uniform sampling of each ray. One point, point The coarse occupancy probability is output by the layered decoder. Fine occupancy probability and color prediction For each ray, coarse depth rendering, fine depth rendering, and color rendering are respectively used... , and It means that, among them, , , ; , .

3. The autonomous parking method based on simultaneous localization and mapping according to claim 2, characterized in that, The process of minimizing the image reconstruction loss in step S2 is as follows: First, create a global keyframe list, and then add new keyframes gradually based on information gain. Next, sample from the current frame and keyframes. 1 pixel and Representing its true depth and color information respectively, and finally calculating and minimizing this... The reconstruction loss per pixel includes geometric loss. and light loss Among them, geometric loss Due to rough geometry loss and fine geometric loss It consists of two parts. , , ; Then, a real-time multi-stage optimization strategy is adopted to minimize the reconstruction loss, first minimizing the coarse geometric loss. To optimize the intermediate layer geometry mesh, and then by minimizing We work together to optimize the intermediate and fine-grained geometries, and finally minimize the total loss. To jointly optimize the geometric mesh, color decoder, and camera extrinsic parameters of K selected keyframes at all levels, where , It is used to balance the loss of light intensity. The hyperparameters of the camera, the camera extrinsic parameters, are used to describe the camera's attitude and position in the world coordinate system, specifically the camera's rotation matrix and translation vector.

4. The autonomous parking method based on simultaneous localization and mapping according to claim 3, characterized in that, The object detection model mentioned in step S3 is a trained YoLo model or a Mask R-CNN model.

5. The autonomous parking method based on simultaneous localization and mapping according to claim 4, characterized in that, The parking space evaluation function mentioned in step S4 is: ,in, , and These represent the size, availability, and distance of the parking space, respectively. , and This is a hyperparameter.

6. The autonomous parking method based on simultaneous localization and mapping according to claim 5, characterized in that, The path planning algorithm described in step S5 uses the RND3QN algorithm.

7. An autonomous parking system based on simultaneous localization and mapping (SLAM), characterized in that, Implementing the method according to any one of claims 1-6, comprising: The one-click parking module includes a one-click parking button, which allows the car owner to initiate the autonomous parking process by pressing the button. The synchronous localization and mapping module uses a dense visual synchronous localization and mapping model based on neural implicit scalable coding to locate the vehicle and build a scene map in real time to accurately reflect environmental information. The vacant parking space detection module uses a target detection model to detect parking spaces and identify vacant parking spaces. The parking space evaluation and optimal parking space selection module selects the target parking space as the optimal parking space based on the parking space evaluation function. The path planning module uses path planning algorithms to determine the optimal driving path and autonomously avoid obstacles. The one-click car-calling module includes a one-click car-calling button, which allows car owners to initiate the autonomous car-calling process by selecting the one-click car-calling button. The self-test module for autonomous vehicles determines whether an autonomous vehicle is usable and sends the results back to the owner. The autonomous vehicle positioning and path planning module receives positioning information sent by the vehicle owner, determines the owner's current location, and uses the same method as the synchronous positioning and mapping module to locate the vehicle and build a scene map. Then, based on the owner's location and the scene map, it performs path planning to determine the optimal path.

Images