An automatic parking method and related device

Abstract

This application belongs to the field of automated parking and relates to an automated parking method and related equipment, including: acquiring environmental data through multiple sensors of the vehicle; mapping the environmental data to a preset vehicle coordinate system after spatiotemporal calibration; separating static and dynamic obstacle features from the environmental data; correcting the initial parking space based on the static and dynamic obstacle features to obtain the target parking space, and tracking and detecting dynamic obstacle behavior; combining the detection results and the target parking space, and outputting parking decisions and path planning through a decision and path planning model. This application achieves accurate separation of static and dynamic obstacle features through multi-sensor fusion and spatiotemporal calibration. Combined with parking space boundary correction and dynamic target tracking and detection, it improves the robustness of obstacle recognition and the accuracy of dynamic target prediction in complex scenarios, optimizes the safety and efficiency of parking strategies, and enhances the scenario adaptability and reliability of the automated parking system.

Description

Technical Field

[0001] This application relates to the field of automatic parking, and more particularly to an automatic parking method and related equipment. Background Technology

[0002] With the development of autonomous driving technology, automatic parking, as a core function to improve driving convenience, has become an important feature of intelligent vehicles. Existing automatic parking systems typically achieve parking space recognition and obstacle detection through a single sensor or a simple fusion scheme. Its basic process includes environmental perception, parking space selection, path planning, and motion control, and can complete basic horizontal and vertical parking functions.

[0003] However, existing technologies have significant drawbacks in complex parking environments. Static obstacle recognition lacks robustness because traditional cameras are susceptible to changes in lighting and blurred road markings, while ultrasonic radar echoes from low obstacles are easily interfered with by clutter. Dynamic obstacle tracking is prone to failure, with camera trajectories lost and radar resolution insufficient in occluded or low-light scenarios. Adaptability to different parking space types is poor, with limited coverage for angled or non-standard parking spaces. Sensor fusion accuracy is low, and fixed-weight schemes do not consider dynamic scene changes, leading to significant positioning errors. These issues render existing systems unreliable in complex scenarios, failing to meet users' demands for safety, efficiency, and scene adaptability. Summary of the Invention

[0004] The purpose of this application is to provide an automatic parking method, device, computer equipment, and storage medium to solve problems such as insufficient reliability of environmental perception, poor robustness of static obstacle recognition, low accuracy of dynamic target tracking, and weak adaptability to non-standard parking spaces caused by a single sensor or simple fusion scheme.

[0005] To address the aforementioned technical problems, this application provides an automatic parking method, employing the following technical solution:

[0006] An automatic parking method includes the following steps:

[0007] The system acquires environmental data detected by multiple sensors on the vehicle, performs spatiotemporal calibration on the environmental data, and generates calibrated environmental data, which includes distance detection data, image data, and motion state data.

[0008] The calibrated environmental data is uniformly mapped to a preset vehicle coordinate system, under which static and dynamic obstacles are identified;

[0009] Obtain the static obstacle features of the static obstacle and the dynamic obstacle features of the dynamic obstacle;

[0010] Extract parking space information from the calibrated environmental data to obtain initial parking spaces;

[0011] Based on the static obstacle features and the dynamic obstacle features, the boundary of the initial parking space is corrected to obtain the target parking space;

[0012] Based on the characteristics of the dynamic obstacle, motion detection is performed on the dynamic obstacle to obtain the detection result;

[0013] Based on the detection results and the target parking space, parking decisions and path planning for the vehicle are generated using a parking decision model and a path planning model, respectively.

[0014] Furthermore, the above-mentioned spatiotemporal calibration processing of the environmental data to generate calibrated environmental data includes:

[0015] The time axis of the image data and the time axis of the distance detection data are synchronized to obtain synchronized image data and synchronized distance detection data; the synchronized image data is calibrated using intrinsic parameters to obtain calibrated image data; the synchronized distance detection data is calibrated using extrinsic parameters to obtain calibrated distance detection data; and the calibrated environmental data is generated based on the calibrated image data and the calibrated distance detection data.

[0016] Furthermore, the acquisition of the static obstacle features of the static obstacle and the dynamic obstacle features of the dynamic obstacle includes:

[0017] The image data is extracted to obtain image features; the distance detection data is filtered to obtain distance features; the dynamic obstacle is analyzed based on the motion state data to obtain motion features; the image features and the distance features are weighted and fused to obtain static obstacle features; the image features, the distance features, and the motion features are weighted and fused to obtain dynamic obstacle features.

[0018] Furthermore, the boundary of the initial parking space is corrected based on the static obstacle features and the dynamic obstacle features to obtain the target parking space, including:

[0019] Based on the static obstacle features and the dynamic obstacle features, the parking space type is determined; based on the parking space type, the boundary of the initial parking space is corrected to obtain the target parking space.

[0020] Furthermore, the determination of parking space type based on the static obstacle characteristics and the dynamic obstacle characteristics includes:

[0021] The static obstacle is analyzed based on its static obstacle characteristics to obtain its category and location information; the dynamic obstacle is detected based on its dynamic obstacle characteristics to obtain its state; and the parking space type is determined based on its category, location information, and state.

[0022] Furthermore, based on the dynamic obstacle characteristics, the above-mentioned motion detection of the dynamic obstacle to obtain the detection result includes:

[0023] The trajectory of the dynamic obstacle within the target field of vision of the vehicle is detected; based on the trajectory, the collision risk between the vehicle and the dynamic obstacle is predicted, and the detection result is obtained.

[0024] Furthermore, the aforementioned sensors include lidar, vision sensors, and millimeter-wave radar. The acquisition of environmental data detected by multiple sensors on the vehicle includes:

[0025] The distance detection data is detected by the lidar, the image data is detected by the vision sensor, and the motion state data is detected by the millimeter-wave radar.

[0026] To address the aforementioned technical problems, this application also provides an automatic parking device, which employs the following technical solution:

[0027] An automatic parking device includes:

[0028] The acquisition module is used to acquire environmental data detected by multiple sensors on the vehicle, perform spatiotemporal calibration processing on the environmental data, and generate calibrated environmental data, which includes distance detection data, image data, and motion state data.

[0029] The mapping module is used to uniformly map the calibrated environmental data to a preset vehicle coordinate system, and to identify static and dynamic obstacles in the vehicle coordinate system.

[0030] The feature module is used to acquire the static obstacle features of the static obstacle and the dynamic obstacle features of the dynamic obstacle;

[0031] The parking space recognition module is used to extract parking space information from the calibrated environmental data to obtain the initial parking space;

[0032] The correction module is used to correct the boundary of the initial parking space based on the static obstacle features and the dynamic obstacle features to obtain the target parking space;

[0033] The detection module is used to perform motion detection on the dynamic obstacle based on its characteristics and obtain the detection result.

[0034] The output module is used to generate parking decisions and path plans for the vehicle based on the detection results and the target parking space, using a parking decision model and a path planning model, respectively.

[0035] To address the aforementioned technical problems, this application also provides a computer device that employs the following technical solution:

[0036] A computer device includes a memory and a processor, the memory storing computer-readable instructions, the processor executing the computer-readable instructions to implement the steps of an automatic parking method as described above.

[0037] To address the aforementioned technical problems, this application also provides a computer-readable storage medium, employing the technical solution described below:

[0038] A computer-readable storage medium storing computer-readable instructions that, when executed by a processor, implement the steps of an automatic parking method as described above.

[0039] Compared with the prior art, the embodiments of this application have the following main advantages:

[0040] This application discloses an automatic parking method that acquires environmental data, including image data, distance detection data, and motion state data, through multiple sensors of the vehicle. After spatiotemporal calibration, the data is uniformly mapped to the vehicle coordinate system to separate static and dynamic obstacle features. Static obstacle features are corrected according to the parking space type to obtain parking space boundary correction results, and dynamic obstacle features are tracked and their behavior detected. Combining the detection results, correction results, and vehicle information, parking decisions and path planning are output through a parking decision model and a path planning model.

[0041] This application achieves precise separation of static and dynamic obstacle features by mapping environmental data to the vehicle coordinate system through multi-sensor fusion and spatiotemporal calibration. Combined with parking space boundary correction and dynamic target tracking and detection, it improves the robustness of obstacle recognition and the accuracy of dynamic target prediction in complex scenarios, optimizes the safety and efficiency of parking strategies, and enhances the scenario adaptability and reliability of the automatic parking system. Attached Figure Description

[0042] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is an exemplary system architecture diagram to which this application can be applied;

[0044] Figure 2 This is a flowchart of an embodiment of an automatic parking method according to this application;

[0045] Figure 3 This is a schematic diagram of a vehicle coordinate system according to an automatic parking method of this application;

[0046] Figure 4 This is a schematic diagram of parking in a parking lot, representing an embodiment of an automatic parking method according to this application.

[0047] Figure 5 This is a schematic diagram of one embodiment of an automatic parking device based on sensor fusion according to this application;

[0048] Figure 6 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0049] Unless otherwise defined, 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 application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0050] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0051] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0052] like Figure 1 As shown, the system architecture 100 may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing communication links between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0053] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0054] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, e-book readers, MP3 (Moving Picture Experts Group Audio Layer III) players, MP4 (Moving Picture Experts Group Audio Layer IV) players, laptops, and desktop computers, etc.

[0055] Server 105 can be a server that provides various services, such as a backend server that supports the pages displayed on the first terminal device 101, the second terminal device 102, and the third terminal device 103.

[0056] It should be noted that the automatic parking method provided in this application is generally executed by a terminal device, and correspondingly, an automatic parking device based on sensor fusion is generally installed in the terminal device.

[0057] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0058] Continue to refer to Figure 2A flowchart of an embodiment of an automatic parking method according to this application is shown. The automatic parking method includes the following steps:

[0059] Step S1: Acquire environmental data detected by multiple sensors on the vehicle, perform spatiotemporal calibration processing on the environmental data, and generate calibrated environmental data, which includes distance detection data, image data, and motion state data.

[0060] In this embodiment, an automatic parking method operates on electronic devices (e.g., Figure 1 The terminal device shown can send or receive data via wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connections, Wi-Fi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultrawideband) connections, and other currently known or future wireless connection methods.

[0061] It's important to note that automatic parking relies on a comprehensive understanding of the surrounding environment, including parking space location, obstacle distribution, and the vehicle's own status such as current position and speed. A single sensor cannot cover all scenarios; for example, cameras are affected by lighting conditions, and radar lacks texture information. Different sensors have complementary physical characteristics. For instance, vision sensors excel at recognizing textures and contours, while distance sensors are adept at detecting position and speed. Therefore, multiple sensors are needed to collect the necessary information for parking, such as location, object type, and motion state. Thus, a variety of sensors must work together to collect environmental data.

[0062] It should be noted that different sensors have different sampling times and installation locations. For example, a camera has a frame rate of 30Hz while a radar has a frame rate of 10Hz. Direct fusion will lead to temporal and spatial misalignment of the data. For instance, the same obstacle may appear in different positions in the image and radar data, affecting the accuracy of subsequent recognition. Therefore, spatiotemporal calibration is necessary to unify the data and calibrate the environmental data in two aspects.

[0063] Environmental data is continuously collected in real time through multiple sensors. This data can be acquired via cameras, scanners, spectral cameras, etc., capturing images and identifying environmental data such as parking lines and obstacle textures. Various radar systems, such as millimeter-wave radar and long-range radar, can also collect environmental data, obtaining information such as the distance, outline, and motion status of obstacles.

[0064] By using time synchronization processing, sampling data from different sensors can be mapped to the same time base to ensure that the data corresponds to the environmental state at the same moment. This can be achieved by using methods such as timestamp alignment based on a unified clock source and interpolation compensation.

[0065] Spatial calibration is used to unify measurement data from different sensors to the same spatial reference (such as the sensor's own coordinate system or a temporary reference coordinate system) through methods such as coordinate transformation and parameter correction, thereby eliminating spatial position deviations. Methods can include coordinate transformation based on sensor installation parameters and error compensation based on calibration.

[0066] Step S2: The calibrated environmental data is uniformly mapped to a preset vehicle coordinate system, and static and dynamic obstacles are identified in the vehicle coordinate system.

[0067] The calibrated sensor data is still based on its own coordinate system. For example, the camera uses the center of the lens as the origin, and the radar uses the installation point as the origin. They cannot be directly fused and processed. They need to be mapped to the same coordinate system to facilitate unified analysis of the relative positions of obstacles and vehicles. Define a coordinate system with the vehicle as the reference, and transform all calibrated environmental data (images, point clouds, radar data) to this coordinate system to achieve spatial unification of the data.

[0068] Specifically, a three-dimensional or two-dimensional coordinate system is defined, with a certain feature point of the vehicle (such as the vehicle's center of gravity, rear axle center, or front bumper midpoint) as the origin and the vehicle's direction of motion as a certain coordinate axis (such as the X-axis). The specific definition of the coordinate system can be flexibly set. Through preset coordinate transformation rules, the vehicle dimensions and sensor installation positions can be transformed into the vehicle coordinate system using transformation matrices and geometrically based conversion formulas, thereby achieving spatial unification of all data.

[0069] Example, reference Figure 3 , Figure 3 This is an example diagram of a vehicle coordinate system. The default vehicle coordinate system is a three-dimensional rectangular coordinate system with the center point of the rear axle of the vehicle as the origin (O), the direction of vehicle movement as the positive X-axis, the direction perpendicular to the direction of movement to the left as the positive Y-axis, and the direction vertically upward as the positive Z-axis.

[0070] Using an extrinsic parameter matrix (a transformation matrix for vehicle dimensions and sensor mounting positions), the calibrated image pixel coordinates (after conversion to the camera coordinate system via intrinsic parameters), LiDAR point cloud coordinates, and millimeter-wave radar detection coordinates are uniformly transformed to the vehicle coordinate system. For example, if the LiDAR detects an obstacle with coordinates (2m, 1m, 0m) in its local coordinate system, after transformation using the extrinsic parameter matrix, its coordinates in the vehicle coordinate system will be (3.5m, 0.8m, 0m), indicating that the obstacle is located 3.5m in front of the vehicle and 0.8m to its left.

[0071] Step S3: Obtain the static obstacle features of the static obstacle and the dynamic obstacle features of the dynamic obstacle.

[0072] The processing objectives for static obstacles (such as pillars and parking locks) and dynamic obstacles (such as pedestrians and moving vehicles) differ. Static obstacles are used to correct parking space boundaries, while dynamic obstacles need to be tracked and predicted to avoid collisions. Mixing these processes can lead to feature confusion (such as misclassifying stationary vehicles as dynamic targets), reducing the accuracy of subsequent decisions; therefore, they must be separated. First, feature analysis is performed on the calibrated environmental data in the vehicle coordinate system, followed by feature extraction.

[0073] Feature analysis distinguishes between static and dynamic obstacles by analyzing the temporal changes in environmental data (such as whether the position of an object changes in multiple consecutive frames of data) and extracting motion parameters (such as whether there is velocity or acceleration).

[0074] For static obstacles, their spatial features (such as shape, size, and position coordinates in the vehicle coordinate system) are extracted to obtain static obstacle features. For dynamic obstacles, their spatial and motion features (such as real-time position, velocity, acceleration, and direction of motion) are extracted to obtain dynamic obstacle features.

[0075] Step S4: Extract parking space information from the calibrated environmental data to obtain the initial parking spaces.

[0076] The goal of automated parking is to safely park a vehicle in a suitable parking space, which requires first locating potential parking spaces from complex environmental data. The raw environmental data contains interference information such as obstacles and invalid areas. Only by extracting targeted parking space information can areas with parking potential be selected.

[0077] Specifically, a general feature model for parking spaces is predefined, including the boundary shape of typical parking spaces (such as straight boundaries and polygonal boundaries), size range (such as minimum thresholds for length and width), and spatial location attributes (such as their relative relationship with the road). Areas matching the predefined parking space feature model are searched in the environmental data. By comparing the boundary contours, size parameters, and surrounding environment (such as the presence of parking line markings), potential parking space areas are identified. For the identified potential areas, key information such as their boundary coordinates in the vehicle coordinate system, actual dimensions, relative distance to vehicles, and orientation is extracted. Areas that meet the "preliminary parking" criteria are marked as initial parking spaces.

[0078] For example, in parking lot environmental data processing, the system searches based on a preset parking space feature model (straight boundary, length ≥ 5m, width ≥ 2.2m, presence of continuous parking space line markers). From the environmental data, it identifies two areas matching these characteristics: one 4m to the left and one 8m above the vehicle. The 4m area has a boundary formed by two parallel straight lines, 5.5m long and 2.3m wide, perpendicular to the vehicle's direction of travel. The other 8m area has a length of 6m and a width of 2.2m, inclined to the vehicle's direction of travel. The system extracts the boundary coordinates and location information of these two areas and marks them as initial parking spaces.

[0079] Step S5: Based on the static obstacle features and the dynamic obstacle features, the boundary of the initial parking space is corrected to obtain the target parking space.

[0080] The initially identified parking space may be unusable due to static obstacles (such as pillars or parking locks), or the space may be wide enough but have a pillar on the left, causing scraping if parked directly. Furthermore, dynamic obstacles (such as pedestrians or shopping carts) may temporarily enter the space; without correction, parking along the original boundaries will result in a collision. By combining the characteristics of static obstacles (such as the location and size of parking locks) and dynamic obstacles (such as moving objects temporarily entering the space), adjustments such as shrinking, shifting, and expanding the initial parking space can be made to avoid risks inherent in the initial parking space.

[0081] Based on the center point coordinates and outline dimensions of static obstacles, it is determined whether they are located within the initial parking space boundary or whether their distance from the initial parking space boundary is less than a safety threshold. For example, if the center point coordinates of a parking lock are within the initial parking space boundary, or if the distance between the post and the initial parking space boundary is 0.2m < 0.3m, it is determined that "it affects the initial parking space". Risk levels are categorized according to the size, material, and avoidability of static obstacles. For example, high risk (e.g., parking locks, posts, unavoidable, requiring boundary contraction), medium risk (e.g., speed bumps, avoidable, requiring local path adjustment), and low risk (e.g., small stones, no impact, no correction required). Boundary correction is only applied to high-risk static obstacles.

[0082] Based on the real-time coordinates and relative motion trends of dynamic obstacles, it is determined whether they are currently within the initial parking space or are likely to enter the initial parking space within a preset time (e.g., within 3 seconds). For example, if a pedestrian's current coordinates are within the initial parking space, or if the trajectory of a moving vehicle, after fitting, shows that it will enter the initial parking space within 3 seconds, it is determined that "it has an impact on the initial parking space".

[0083] If a static obstacle is located within the initial parking space, the parking space boundary is reduced according to the obstacle's outline dimensions and a safety margin (usually 0.1–0.2 m). If a dynamic obstacle currently occupies a portion of the initial parking space and its impact lasts for an extended period (e.g., 10 seconds), that portion is trimmed. When both static and dynamic obstacles affect the initial parking space, the rules for high-risk obstacles (such as pedestrians and vehicles) take precedence.

[0084] According to the correction rules, the coordinates of the four vertices of the initial parking space are recalculated. For example, if the right boundary of the initial parking space is Y1.2m, it needs to be reduced to Y0.95m due to the influence of the parking lock. Therefore, the coordinates of the two vertices of the right boundary are corrected from (X2.0, Y1.2) and (X7.2, Y1.2) to (X2.0, Y0.95) and (X7.2, Y0.95). After correction, it is verified whether the length and width of the target parking space still meet the minimum requirements for vehicle parking (length ≥ vehicle length + 0.3m, width ≥ vehicle width + 0.3m). If they meet the requirements, continue; if they do not meet the requirements, the initial parking space is determined to be unusable, and the process returns to step S4 to extract other initial parking spaces. It is confirmed that there are "no high-risk static obstacles and no dynamic obstacles that have been occupied for a long time" within the corrected boundary to ensure the safety of the target parking space. Finally, the corrected boundary coordinates, center point coordinates, actual dimensions, and risk labels (such as 'no high-risk obstacles' and 'pay attention to dynamic pedestrians on the right') are integrated to form the target parking space.

[0085] Step S6: Based on the characteristics of the dynamic obstacle, perform motion detection on the dynamic obstacle to obtain the detection result.

[0086] The movement of dynamic obstacles is random. For example, a pedestrian may suddenly cross the road or a vehicle may drive out. If only real-time detection is performed without motion prediction, a collision may occur due to reaction lag. Therefore, it is necessary to track their trajectory and predict their behavior, calculate the time to collision (TTC), assess the risk in advance, and output the detection results.

[0087] By comparing the temporal data of continuous frames of environmental data, the real-time speed and direction of dynamic obstacles relative to vehicles are calculated to determine whether their movement is uniform and whether there are sudden speed changes (such as sudden acceleration or deceleration). For example, speed data is updated every 0.1 seconds. If the difference between two adjacent speeds is greater than 0.5 m / s, it is judged as a speed change and requires close attention. Based on the historical position coordinates of dynamic obstacles (such as data from the past 10 frames), their movement trajectory (such as straight lines or curves) is fitted, and the trend of position changes in the near future is predicted to determine whether their movement direction is towards the parking path and whether they will enter the safe area around the parking space (such as within 1 meter in front of the parking space). For example, by fitting the trajectory using a polynomial, the position of a pedestrian 1 second later is predicted. If the predicted position is on the parking path, it needs to be marked as high risk. Combining the behavioral characteristics of dynamic obstacles, their movement patterns (such as uniform straight line, variable speed steering, and paused waiting) are identified. For example, a moving vehicle with its turn signal on and speed decreasing is identified as a "preparing to turn" mode; a pedestrian's speed decreasing to 0 and their posture standing is identified as a "paused waiting" mode. Different modes correspond to different collision risk levels.

[0088] Based on motion state detection results, the collision risk between dynamic obstacles and vehicles is assessed. The time to collision (TTC) is calculated if the obstacle and vehicle continue along their current trajectories in their current motion state. TTC = current straight-line distance between the two obstacles / relative velocity (if the relative velocity is 0, TTC is infinite, and there is no collision risk). For example, if the vehicle and pedestrian are currently 5m apart and their relative velocity is 2.5m / s (the pedestrian is moving towards the vehicle), then TTC = 5 / 2.5 = 2s. The probability of a collision is assessed by comprehensively considering the uncertainty of the obstacle's motion (e.g., the pedestrian may suddenly change direction) and the possibility of adjusting the vehicle's parking path. For example, if the pedestrian's trajectory is unstable (multiple changes of direction), the probability of a collision increases; if the vehicle has a large space for path adjustment (e.g., it can shift laterally), the probability of a collision decreases.

[0089] Risk levels are determined based on TTC and collision probability. For example, high risk is defined as TTC < 2s and collision probability > 60%, or the dynamic obstacle has entered the core area of ​​the parking path (e.g., within 0.5m directly in front of the parking space); medium risk is defined as 2s ≤ TTC < 4s and 30% ≤ collision probability ≤ 60%, or the dynamic obstacle is close to the edge of the parking path; low risk is defined as TTC ≥ 4s and collision probability < 30%, or the dynamic obstacle is far from the parking path.

[0090] For example, during parking in a parking lot, a unique identifier is assigned to a moving target through continuous frame contour similarity comparison, and its movement trajectory is tracked in real time as "moving from 3m to the left of the vehicle towards the parking space". When the target is obscured by a pillar and out of the vehicle's view, an estimated trajectory is generated based on the velocity (0.8m / s) and direction of the previous 5 frames during the obstruction period. Combining the estimated trajectory with the vehicle's preset parking path, the collision time is calculated to be 3.5s, which is classified as medium risk. The output includes a detection result containing "target identifier, estimated trajectory, and medium risk".

[0091] Step S7: Based on the detection results, the correction results, and the vehicle information, output parking decisions and path plans respectively through the parking decision model and the path planning model.

[0092] Parking decisions (whether to park) and route planning (how to park) need to take into account environmental constraints, such as the revised parking space boundaries, dynamic obstacle risks and vehicle status, vehicle size, minimum turning radius, etc. Otherwise, an unavailable parking space may be selected or a dangerous route may be planned.

[0093] It should be noted that the parking decision model can be composed of models such as TOPSIS (Technique for Order Preference by Similarity to an Ideal Solution), BP neural network (backpropagation), or a fusion model of ResNet and BERT.

[0094] For example, a rule-based weighted decision model is used. The inputs are the dynamic obstacle detection results (low collision risk level, no spatial overlap, TTC=6s) and the target parking space information (vertical type, available, 4.5m away from the current position of the vehicle). The model calculates the weights according to the rules (0.6 for low risk, 0.4 for close and available parking space, and 1.0 for overall weight) and outputs the parking decision "Continue parking, driving speed 1.2m / s, monitor dynamic obstacles in real time during the reversing process, and immediately decelerate if the risk level increases".

[0095] It should be noted that the path planning model can be a combination of the Rapidly-exploring Random Tree (RRT) model and the quadratic programming local adjustment model, or it can be a model such as the Long Short-Term Memory (LSTM) network.

[0096] For example, a simplified RRT path planning model is used. The inputs are the vehicle's current position (X0,Y0), the vehicle's minimum turning radius of 3.5m, the center point of the target parking space (X4.6,Y-0.125) and its boundaries (X2.0-7.2m,Y-1.2~0.95m), and static obstacles (pillar X5.0,Y1.5). The model searches and generates a path plan: the path point sequence is (X0,Y0) to (X1.0,Y0) to (X2.5,Y-0.3) to (X3.8,Y-0.2) to (X4.6,Y-0.125), the corresponding turning angles are 0° to 3° to 12° to 5° to 0°, and the driving speeds are 1.2m / s to 1.0m / s to 0.8m / s to 0.5m / s to 0m / s (stopping after reaching the parking space).

[0097] This application utilizes multi-sensor fusion and spatiotemporal calibration to uniformly map environmental data to the vehicle coordinate system, achieving precise separation of static and dynamic obstacle features. Combined with parking space boundary correction and dynamic target tracking and detection, it improves the robustness of obstacle recognition and the accuracy of dynamic target prediction in complex scenarios, optimizes the safety and efficiency of parking strategies, and enhances the scenario adaptability and reliability of the automatic parking system.

[0098] In some optional implementations of this embodiment, the above-described spatiotemporal calibration processing of the environmental data to generate calibrated environmental data includes:

[0099] The time axis of the image data and the time axis of the distance detection data are synchronized to obtain synchronized image data and synchronized distance detection data; the synchronized image data is calibrated using intrinsic parameters to obtain calibrated image data; the synchronized distance detection data is calibrated using extrinsic parameters to obtain calibrated distance detection data; and the calibrated environmental data is generated based on the calibrated image data and the calibrated distance detection data.

[0100] In this embodiment, the timestamp of the vehicle CAN bus is used as a reference (accuracy ≤1ms) to match the timestamps of image data and distance detection data. Data loss is compensated by interpolation algorithm, and the time synchronization error is controlled within ≤5ms.

[0101] Intrinsic parameter calibration corrects image distortion by calculating distortion coefficients (such as radial distortion k1, k2) and focal lengths (fx, fy) using chessboard-patterned images. Extrinsic parameter calibration aligns the LiDAR and camera with the same calibration board (including corner points and reflectors), and calculates the extrinsic parameter matrix (rotation matrix R and translation vector T) by matching point cloud with image features, thus aligning the LiDAR point cloud coordinates with the camera pixel coordinates. The extrinsic parameters of millimeter-wave radar and LiDAR are calibrated similarly.

[0102] For example, a vehicle's camera and LiDAR have an installation misalignment. Before time synchronization, the camera captures an image frame at t=10.000s, while the LiDAR captures a point cloud at t=10.003s. The 3ms time difference causes a mismatch between "pedestrian at position A" in the image and "pedestrian at position B" in the point cloud. After time synchronization, interpolation is used to map the LiDAR data to t=10.000s, ensuring that the pedestrian's position corresponds in both the image and the point cloud.

[0103] Before spatial calibration, the coordinates of the pillar detected by the lidar in its local coordinate system were (2.0m, 1.0m), while the corresponding coordinates of the same pillar detected by the camera in the image, after conversion, were (2.3m, 1.2m), a deviation of 0.3m. After spatial calibration, through extrinsic parameter matrix transformation, the deviation between the lidar data and the camera data was reduced to 2cm, achieving uniformity in the position of the same obstacle. The calibrated image data and the calibrated distance detection data were then integrated into calibrated environmental data.

[0104] This application eliminates spatiotemporal bias by synchronizing and calibrating the data in time, ensuring that the description of the same environment by multi-source data remains consistent, and providing a reliable foundation for subsequent feature fusion and obstacle recognition.

[0105] In some optional implementations of this embodiment, obtaining the static obstacle features of the static obstacle and the dynamic obstacle features of the dynamic obstacle includes:

[0106] The image data is extracted to obtain image features; the distance detection data is filtered to obtain distance features; the dynamic obstacle is analyzed based on the motion state data to obtain motion features; the image features and the distance features are weighted and fused to obtain static obstacle features; the image features, the distance features, and the motion features are weighted and fused to obtain dynamic obstacle features.

[0107] By analyzing visual information to extract image features and using computer vision algorithms, visual features related to obstacles are extracted from the calibrated image, covering three categories: geometric features, texture features, and semantic features. The calibrated image is then converted to grayscale and denoised (Gaussian filtering, kernel size 3×3) to enhance the contrast between obstacles and the background.

[0108] Static obstacles can be detected by improving the Canny operator (with adaptive threshold adjustment, reducing high thresholds in low light), and then using a contour extraction algorithm (such as findContours) to obtain the obstacle contour, calculating the bounding rectangle (length and width) and roundness (to determine if it is a pillar). Dynamic obstacles can be detected by calculating pixel displacement using optical flow; areas with pixel displacement > 10px are identified as moving regions, and the outer rectangle size of the moving region is extracted.

[0109] Texture features of static and dynamic obstacles are extracted. For static obstacles (such as parking locks), the Local Binary Patterns (LBP) algorithm is used to extract texture features (e.g., the embossed pattern of the parking lock), generating a 64-dimensional texture feature vector. For dynamic obstacles (such as pedestrians), the Histogram of Oriented Gradients (HOG) algorithm is used to extract contour texture, generating a 3780-dimensional HOG feature vector. Geometric features (e.g., 0.3m long and 0.2m wide) are concatenated with the texture feature vector to form an image feature set.

[0110] Distance features can be obtained by extracting distance information and then filtering and analyzing it. The calibrated distance detection data (such as radar point clouds and ultrasonic echoes) is filtered and denoised to extract features such as the distance, size, and relative position of obstacles, eliminating static clutter and measurement noise.

[0111] For ultrasonic radar data, retain data with distance abrupt changes ≤0.3m and echo intensity >80 (such as ground lock echoes), and filter out static clutter (such as ground reflections) with stable distances and echo intensity <50. For lidar data, use statistical filtering (removing isolated points whose distance from neighboring points is greater than 2 standard deviations) to remove noise points and retain obstacle point clouds.

[0112] Calculate the minimum distance between the obstacle and the vehicle (e.g., 7.5m from the vehicle to the ground lock) and the rate of change of distance (≤0.01m / s for static obstacles, >0.2m / s for dynamic obstacles) to obtain distance features. Cluster the filtered point cloud (e.g., Euclidean clustering, distance threshold 0.2m) and calculate the dimensions of the circumscribed cube of the clustered obstacles (e.g., 0.5m diameter, 2.5m height for a pillar) to obtain size features. Integrate the distance, rate of change of distance, and size features to form a distance feature set.

[0113] Motion features can be extracted through dynamic obstacle motion trend analysis. Based on calibrated motion state data (such as vehicle speed, steering angle, and relative speed of obstacles), the motion features of dynamic obstacles, such as speed, acceleration, and direction of motion, are obtained through time-series analysis.

[0114] Acquire dynamic obstacle position data (X / Y coordinates in the vehicle coordinate system) and vehicle speed (e.g., 1 km / h) for 5 consecutive frames (0.1 s interval between each frame). Based on the change in obstacle position ΔS between adjacent frames, divide by the time interval ΔT (0.1 s) to obtain the relative velocity v = ΔS / ΔT (e.g., 0.8 m / s for a pedestrian, 1.2 m / s for a moving vehicle). Based on the change in relative velocity Δv between consecutive frames, divide by the time interval ΔT to obtain the acceleration a = Δv / ΔT (e.g., 0 m / s² for a pedestrian moving at a constant speed, 0.5 m / s² for an accelerating vehicle). Determine the direction of motion based on the sign of the X / Y coordinate changes (e.g., ΔX > 0, ΔY = 0, indicating movement along the positive X-axis of the vehicle coordinate system). Integrate the relative velocity, acceleration, and direction of motion to form a motion feature set.

[0115] The fusion of static obstacle features is achieved through a weighted fusion of image features and distance features. Based on the reliability of the sensor in different scenarios, dynamic weights are assigned to image features and distance features, and a comprehensive feature is generated by weighted summation, thereby improving the robustness of static obstacle features.

[0116] A preset weighting strategy is used: camera geometric features have a weight of 0.6, radar distance features have a weight of 0.4 (in well-lit scenes), and in low-light scenes (image sensor confidence < 0.5), the radar distance feature weight is increased to 0.6, while the camera feature weight is reduced to 0.4. Image features (e.g., contour width 0.2m) and distance features (e.g., radar detection width 0.22m) are normalized to the [0,1] interval to avoid the influence of dimensional differences on the fusion result.

[0117] The standardized features are summed according to their weights. For example, the contour width fusion value = 0.2 × 0.6 + 0.22 × 0.4 = 0.12 + 0.088 = 0.208. The distance fusion value = 7.5 (image estimated distance) × 0.6 + 7.53 (radar detection distance) × 0.4 = 4.5 + 3.012 = 7.512m. The fused width, distance, and texture features are then concatenated to generate a static obstacle feature vector (e.g., [0.208m, 7.512m, LBP texture vector]).

[0118] The fusion of dynamic obstacle features involves weighted fusion of image features, distance features, and motion features. Based on static feature fusion, motion features are added, and their weights are adjusted according to the speed of the dynamic obstacle, prioritizing the retention of high-reliability features (such as radar motion features for high-speed targets).

[0119] A preset weighting strategy is used: for dynamic targets with speeds >1 m / s (e.g., moving vehicles), the radar motion feature weight is set to 0.7, the image feature weight to 0.2, and the distance feature weight to 0.1 (for more reliable radar speed measurement in high-speed scenes). For targets with speeds ≤1 m / s (e.g., pedestrians), the camera visual feature weight is set to 0.6, the distance feature weight to 0.2, and the motion feature weight to 0.2 (for more accurate visual contour recognition in low-speed scenes). Image features (e.g., HOG contour features), distance features (e.g., relative distance of 8 m), and motion features (e.g., speed of 0.8 m / s) are normalized to the [0,1] interval.

[0120] Finally, weighted fusion is performed. For example, for a pedestrian (speed 0.8m / s ≤ 1m / s), the contour confidence fusion value = 0.9 (image HOG confidence) × 0.6 + 0.85 (radar clustering confidence) × 0.2 + 0.8 (motion continuity confidence) × 0.2 = 0.54 + 0.17 + 0.16 = 0.87. The speed fusion value = 0.8 (image optical flow estimated speed) × 0.6 + 0.82 (radar measurement speed) × 0.2 + 0.8 (temporal calculation speed) × 0.2 = 0.48 + 0.164 + 0.16 = 0.804m / s. The fused contour confidence, speed, distance, and other features are then concatenated to generate a dynamic obstacle feature vector (e.g., [0.87, 0.804m / s, 8m]).

[0121] This application reduces the confusion rate between static and dynamic obstacles by separating their features, providing accurate input for subsequent processing. Through dynamic weight adjustment, it maintains separation stability even in low-light or occluded scenes.

[0122] In some optional implementations of this embodiment, the above-mentioned correction of the boundary of the initial parking space based on the static obstacle features and the dynamic obstacle features to obtain the target parking space includes:

[0123] Based on the static obstacle features and the dynamic obstacle features, the parking space type is determined; based on the parking space type, the boundary of the initial parking space is corrected to obtain the target parking space.

[0124] It's important to note that the correction methods differ for different parking space types. For example, in horizontal parking spaces, the parking direction is forward and backward along the X-axis, and obstacles (such as parking locks) primarily affect the left and right boundaries (requiring space to be left for door opening). In perpendicular parking spaces, the parking direction is left and right along the Y-axis, and obstacles primarily affect the front and rear boundaries (requiring space to be left for reversing). If the boundaries are not corrected according to the parking space type, boundaries that need correction will not be corrected, and boundaries that do not need correction will be excessively contracted. Therefore, it is necessary to first determine the parking space type and then correct the boundaries to ensure that the corrected parking space boundaries accurately adapt to different parking scenarios and improve parking space usability.

[0125] The static obstacle layout (such as the relative positions of adjacent vehicles and pillars) and the impact of dynamic obstacles (such as whether they intersect the parking path) of different parking space types exhibit fixed patterns. By analyzing the characteristics of static and dynamic obstacles, feature templates for parking space types are matched to confirm the parking space type. Based on the static obstacle features (category, location), key parameters are extracted to identify boundaries, resulting in parking space templates, which can include horizontal, vertical, and angled parking space templates. If the trajectory of a dynamic obstacle (such as a pedestrian) is parallel to a line connecting a set of static obstacles (such as parallel to a line connecting left and right vehicles, consistent with the parking direction of a horizontal parking space), the parking space type is further verified. The similarity between the current obstacle layout and various parking space templates is calculated (such as the deviation between the angle of the static obstacle line and the template angle; a deviation of <5° indicates high similarity), and the type with the highest similarity is selected as the confirmed parking space type.

[0126] For different parking space types and risk areas, the boundaries of the initial parking spaces are shrunk, offset, and locally trimmed based on obstacle characteristics (such as the size of static obstacles and the temporary position of dynamic obstacles) to ensure that the corrected boundaries avoid all risk areas and meet the minimum size requirements for vehicle parking (such as length ≥ vehicle length + 0.5m, width ≥ vehicle width + 0.3m).

[0127] This application improves the accuracy of parking space type matching by using obstacle feature matching templates, avoiding incorrect correction directions due to type misjudgment (e.g., correcting a horizontal parking space as a vertical parking space). Simultaneously, it corrects for core risk areas of different parking space types (focusing on left and right for horizontal spaces, and front and back for vertical spaces), improving space utilization. Furthermore, it enhances parking safety by correcting boundaries based on the temporary positions of dynamic obstacles (e.g., avoiding temporarily stopped pedestrians).

[0128] In some optional implementations of this embodiment, determining the parking space type based on the static obstacle features and the dynamic obstacle features includes:

[0129] The static obstacle is analyzed based on its static obstacle characteristics to obtain its category and location information; the dynamic obstacle is detected based on its dynamic obstacle characteristics to obtain its state; and the parking space type is determined based on its category, location information, and state.

[0130] Static obstacle analysis can be achieved by constructing a lightweight classification model (such as MobileNetV3-Lite). The input is a static obstacle feature vector (geometric, distance, and texture features), and the trained neural network outputs the obstacle category (parking lock, post, stationary vehicle, curb, etc.). Simultaneously, based on the distance features in the vehicle coordinate system, the precise coordinates (X, Y, Z) of the obstacle are determined.

[0131] The system obtains the coordinates of the initial parking space, calculates the included angles of adjacent boundaries (90° for horizontal parking spaces, 180° for vertical parking spaces, and 30-60° for angled parking spaces) and the boundary length / width, and uses a feature database of common parking space types to match and confirm the parking space type. Simultaneously, it calculates the minimum distance between obstacles and the initial boundary, taking into account obstacle locations. Differentiated correction rules are established for different parking space types. For example, the correction rule for horizontal parking spaces is: "If the distance between the ground lock / curb and the initial boundary is less than the safety threshold (e.g., 0.4m), shrink the boundary inward by the amount of contraction = safety threshold - actual distance." The correction rule for vertical parking spaces is: "If the space is occupied by front or rear pillars / stationary vehicles, adjust the initial length by the amount of correction = vehicle length + safety clearance (e.g., 0.6m)."

[0132] The state of dynamic obstacles (such as whether they occupy the area enclosed by the initial parking space or cut into a potential parking path) determines the availability of the current parking space type. By analyzing the characteristics of dynamic obstacles, auxiliary information for confirming the parking space type can be extracted (such as whether it affects the validity of static constraint boundaries). Based on the distance features (relative distance) and motion features (speed, trajectory curvature) in the dynamic obstacle features, the real-time position coordinates of the dynamic obstacle can be calculated to predict whether it will stop in the target parking space. By combining the rules with the state of the dynamic obstacle, the parking space type is finally determined.

[0133] This application initially identifies parking space types through static obstacle layout, improving the success rate of parking space type identification in scenarios where parking lines are worn or obstructed, and is suitable for scenarios without clear visual markings, such as old parking lots and temporary open spaces. It also determines the validity of the parking space type by assessing the status of dynamic obstacles (e.g., parking spaces occupied by high-impact dynamic obstacles are temporarily excluded), reducing the selection of invalid parking spaces and improving the efficiency of automatic parking.

[0134] In some optional implementations of this embodiment, the above-mentioned motion detection of the dynamic obstacle based on the dynamic obstacle characteristics to obtain the detection result includes:

[0135] The trajectory of the dynamic obstacle within the target field of vision of the vehicle is detected; based on the trajectory, the collision risk between the vehicle and the dynamic obstacle is predicted, and the detection result is obtained.

[0136] In this embodiment, the tracking of dynamic obstacles can employ the Hungarian algorithm to match dynamic targets in consecutive frames, using feature IOU (≥0.5) and cosine similarity (≥0.7) as criteria to maintain ID continuity (switching rate <5%). For example, if pedestrian A is located at (3m, 1m) in the first frame and partially visible in the second frame due to being occluded by a pillar, he / she is still marked as ID=1 through feature matching.

[0137] When the target is completely occluded (lasting ≤2s), based on a Long Short-Term Memory (LSTM) network, the position during the occlusion period is predicted (error ≤20cm) using the trajectory of the previous 5 frames (0.5s / frame). If a pedestrian is occluded for 1s, the predicted position at the end of the occlusion is (2m, 0.8m), which deviates from the actual position by 15cm. The intersection point with the vehicle parking path is calculated based on the predicted trajectory. The risk level is output using the formula TTC = straight-line distance between obstacle and vehicle / relative speed: TTC < 2s is high risk, 2s ≤ TTC < 4s is medium risk, and TTC ≥ 4s is low risk.

[0138] This application uses trajectory prediction and collision risk calculation to output the risk level in advance, allowing time for adjustment in parking decisions, improving the success rate of dynamic target tracking in occluded scenarios, and avoiding target loss.

[0139] In some optional implementations of this embodiment, the sensors include lidar, vision sensors, and millimeter-wave radar. The acquisition of environmental data detected by multiple sensors on the vehicle includes:

[0140] The distance detection data is detected by the lidar, the image data is detected by the vision sensor, and the motion state data is detected by the millimeter-wave radar.

[0141] Based on the vehicle's perception requirements, select 1-2 LiDAR sensors, preferably deployed on the roof or front and rear bumpers. For vision sensors, preferably four surround-view cameras, deployed around the vehicle body. For millimeter-wave radar, preferably 4-6 sensors, deployed on the front and rear bumpers and door sides. Distribute them at preset intervals to ensure the detection range covers near-range (≤2m), medium-range (2-5m), and long-range (5-10m) areas, forming a multi-layered perception network.

[0142] LiDAR scans the surrounding environment by emitting laser beams. It calculates the relative distance between the target and the sensor by multiplying the time difference between the emitted and reflected laser echoes by the speed of light (2), generating distance detection data (including target coordinates and distance values). Visual sensors convert light signals into electrical signals through a lens imaging element, generating continuous frame image data (including visual features such as parking lines, obstacle outlines, and textures). Millimeter-wave radar emits millimeter-wave electromagnetic waves. By analyzing the Doppler frequency shift of the echoes, it calculates the target's velocity and combines this with the echo delay to calculate the distance, generating motion state data (including target velocity, acceleration, and relative position).

[0143] For example, a vehicle's sensors include 4 vision sensors (1280×720 resolution, 30Hz frame rate, covering 360° of the vehicle's field of view), a 16-line LiDAR (detection range 0-15m, frame rate 15Hz, point cloud resolution 0.1m×0.1m), 4 24GHz millimeter-wave radars (detection range 0-10m, frame rate 10Hz, speed measurement range ±5m / s), and onboard CAN bus sensors (collecting vehicle speed, steering angle, etc.).

[0144] After the vehicle enters the parking lot, during the data acquisition process, the visual sensor captures images of the surroundings in real time, the lidar scans the three-dimensional point cloud, the millimeter-wave radar monitors dynamic targets, and the on-board sensors record the vehicle's motion parameters simultaneously. All data is transmitted to the processing unit via the on-board Ethernet.

[0145] This application compensates for the deficiencies in texture recognition by using a variety of different sensors, provides high-precision geometric information, ensures dynamic target detection in harsh environments (such as heavy rain and obstruction), and improves the integrity of environmental data.

[0146] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0147] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0148] Further reference Figure 5 As a response to the above Figure 2 The implementation of the method shown in this application provides an embodiment of an automatic parking device based on sensor fusion, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0149] like Figure 5 As shown, the automatic parking device 500 described in this embodiment includes: a data acquisition module 501, a mapping module 502, a feature module 503, a parking space recognition module 504, a correction module 505, a detection module 506, and an output module 507. Wherein:

[0150] The acquisition module 501 is used to acquire environmental data detected by multiple sensors on the vehicle, perform spatiotemporal calibration processing on the environmental data, and generate calibrated environmental data. The environmental data includes distance detection data, image data, and motion state data.

[0151] The mapping module 502 is used to uniformly map the calibrated environmental data to a preset vehicle coordinate system, and to identify static and dynamic obstacles in the vehicle coordinate system.

[0152] Feature module 503 is used to acquire the static obstacle features of the static obstacle and the dynamic obstacle features of the dynamic obstacle;

[0153] Parking space recognition module 504 is used to extract parking space information from the calibrated environmental data to obtain initial parking spaces;

[0154] The correction module 505 is used to correct the boundary of the initial parking space based on the static obstacle features and the dynamic obstacle features to obtain the target parking space.

[0155] The detection module 506 is used to perform motion detection on the dynamic obstacle based on the characteristics of the dynamic obstacle and obtain the detection result;

[0156] The output module 507 is used to generate parking decisions and path plans for the vehicle based on the detection results and the target parking space, using a parking decision model and a path planning model, respectively.

[0157] In some embodiments of this application, the acquisition module 501 is further configured to synchronize the time axis of the image data and the time axis of the distance detection data to obtain synchronized image data and synchronized distance detection data; perform intrinsic parameter calibration on the synchronized image data to obtain calibrated image data; perform extrinsic parameter calibration on the synchronized distance detection data to obtain calibrated distance detection data; and generate calibrated environmental data based on the calibrated image data and the calibrated distance detection data.

[0158] In some embodiments of this application, the feature module 503 is further configured to extract features from the image data to obtain image features; filter the distance detection data to obtain distance features; perform motion analysis on the dynamic obstacle based on the motion state data to obtain motion features; perform weighted fusion based on the image features and the distance features to obtain static obstacle features; and perform weighted fusion based on the image features, the distance features, and the motion features to obtain dynamic obstacle features.

[0159] In some embodiments of this application, the correction module 505 is further configured to determine the parking space type based on the static obstacle features and the dynamic obstacle features; and to perform boundary correction on the initial parking space based on the parking space type to obtain the target parking space.

[0160] In some embodiments of this application, the correction module 505 is further configured to analyze the static obstacle based on the static obstacle characteristics to obtain the category and location information of the static obstacle; detect the dynamic obstacle based on the dynamic obstacle characteristics to obtain the state of the dynamic obstacle; and determine the parking space type based on the category, the location information and the state of the dynamic obstacle.

[0161] In some embodiments of this application, the detection module 506 is further configured to detect the trajectory movement result of the dynamic obstacle within the target field of vision of the vehicle; and based on the trajectory movement result, predict the collision risk between the vehicle and the dynamic obstacle to obtain the detection result.

[0162] This application provides an automatic parking device that uses multi-sensor fusion and spatiotemporal calibration to uniformly map environmental data to the vehicle coordinate system, achieving accurate separation of static and dynamic obstacle features. Combined with parking space boundary correction and dynamic target tracking and detection, it improves the robustness of obstacle recognition and the accuracy of dynamic target prediction in complex scenarios, optimizes the safety and efficiency of parking strategies, and enhances the scenario adaptability and reliability of the automatic parking system.

[0163] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0164] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected via a system bus. It should be noted that only the computer device 4 with components 41-43 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0165] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0166] The memory 41 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 41 may be an internal storage unit of the computer device 4, such as the hard disk or memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the computer device 4. Of course, the memory 41 may also include both the internal storage unit and its external storage device of the computer device 4. In this embodiment, the memory 41 is typically used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions for an automatic parking method. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or will be output.

[0167] In some embodiments, the processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to execute computer-readable instructions stored in the memory 41 or to process data, for example, to execute computer-readable instructions for an automatic parking method.

[0168] The network interface 43 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 4 and other electronic devices.

[0169] The computer equipment provided in this application maps environmental data to the vehicle coordinate system in a unified manner through multi-sensor fusion and spatiotemporal calibration, thereby achieving accurate separation of static and dynamic obstacle features. Combined with parking space boundary correction and dynamic target tracking and detection, it improves the robustness of obstacle recognition and the accuracy of dynamic target prediction in complex scenarios, optimizes the safety and efficiency of parking strategies, and enhances the scenario adaptability and reliability of the automatic parking system.

[0170] This application also provides another embodiment, namely, a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of an automatic parking method as described above.

[0171] The computer-readable storage medium provided in this application maps environmental data to the vehicle coordinate system in a unified manner through multi-sensor fusion and spatiotemporal calibration, thereby achieving accurate separation of static and dynamic obstacle features. Combined with parking space boundary correction and dynamic target tracking and detection, it improves the robustness of obstacle recognition and the accuracy of dynamic target prediction in complex scenarios, optimizes the safety and efficiency of parking strategies, and enhances the scenario adaptability and reliability of the automatic parking system.

[0172] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0173] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. An automatic parking method, characterized in that, Includes the following steps: The system acquires environmental data detected by multiple sensors on the vehicle, performs spatiotemporal calibration on the environmental data, and generates calibrated environmental data, which includes distance detection data, image data, and motion state data. The calibrated environmental data is uniformly mapped to a preset vehicle coordinate system, under which static and dynamic obstacles are identified; Obtain the static obstacle features of the static obstacle and the dynamic obstacle features of the dynamic obstacle; Extract parking space information from the calibrated environmental data to obtain initial parking spaces; Based on the static obstacle features and the dynamic obstacle features, the boundary of the initial parking space is corrected to obtain the target parking space; Based on the characteristics of the dynamic obstacle, motion detection is performed on the dynamic obstacle to obtain the detection result; Based on the detection results and the target parking space, the parking decision and path planning for the vehicle are generated respectively through the parking decision model and the path planning model. Wherein, acquiring the static obstacle features of the static obstacle and the dynamic obstacle features of the dynamic obstacle includes: Extract the features from the image data to obtain image features; The distance detection data is filtered to obtain distance features; Based on the motion state data, motion analysis is performed on the dynamic obstacle to obtain its motion characteristics; The static obstacle features are obtained by weighted fusion of the image features and the distance features. The dynamic obstacle features are obtained by weighted fusion of the image features, distance features, and motion features.

2. The automatic parking method according to claim 1, characterized in that, The process of performing spatiotemporal calibration on the environmental data to generate calibrated environmental data includes: Synchronize the time axis of the image data and the time axis of the distance detection data to obtain synchronized image data and synchronized distance detection data; The synchronized image data is subjected to intrinsic parameter calibration to obtain calibrated image data; The synchronized distance detection data is calibrated using external parameters to obtain calibrated distance detection data; The calibrated environmental data is generated based on the calibrated image data and the calibrated distance detection data.

3. The automatic parking method according to claim 1, characterized in that, The step of correcting the boundary of the initial parking space based on the static obstacle features and the dynamic obstacle features to obtain the target parking space includes: Based on the static obstacle characteristics and the dynamic obstacle characteristics, the parking space type is determined; Based on the parking space type, the initial parking space is boundary-corrected to obtain the target parking space.

4. The automatic parking method according to claim 3, characterized in that, The step of determining the parking space type based on the static obstacle characteristics and the dynamic obstacle characteristics includes: The static obstacles are analyzed based on their characteristics to obtain their category and location information. The dynamic obstacle is detected based on its characteristics to obtain its state. The parking space type is determined based on the category, the location information, and the state of the dynamic obstacle.

5. The automatic parking method according to claim 1, characterized in that, The step of performing motion detection on the dynamic obstacle based on its characteristics to obtain detection results includes: The trajectory movement of the dynamic obstacle within the target field of view of the vehicle is detected; Based on the trajectory motion results, the collision risk between the vehicle and the dynamic obstacle is predicted, and the detection result is obtained.

6. The automatic parking method according to claim 1, characterized in that, The sensors include lidar, vision sensors, and millimeter-wave radar. Acquiring environmental data detected by multiple sensors on the vehicle includes: The distance detection data is detected by the lidar, the image data is detected by the vision sensor, and the motion state data is detected by the millimeter-wave radar.

7. An automatic parking device based on sensor fusion, characterized in that, include: The acquisition module is used to acquire environmental data detected by multiple sensors on the vehicle, perform spatiotemporal calibration processing on the environmental data, and generate calibrated environmental data, which includes distance detection data, image data, and motion state data. The mapping module is used to uniformly map the calibrated environmental data to a preset vehicle coordinate system, and to identify static and dynamic obstacles in the vehicle coordinate system. The feature module is used to acquire the static obstacle features of the static obstacle and the dynamic obstacle features of the dynamic obstacle; The parking space recognition module is used to extract parking space information from the calibrated environmental data to obtain the initial parking space; The correction module is used to correct the boundary of the initial parking space based on the static obstacle features and the dynamic obstacle features to obtain the target parking space; The detection module is used to perform motion detection on the dynamic obstacle based on its characteristics and obtain the detection result. The output module is used to generate parking decisions and path plans for the vehicle based on the detection results and the target parking space, using a parking decision model and a path planning model, respectively. Wherein, acquiring the static obstacle features of the static obstacle and the dynamic obstacle features of the dynamic obstacle includes: Extract the features from the image data to obtain image features; The distance detection data is filtered to obtain distance features; Based on the motion state data, motion analysis is performed on the dynamic obstacle to obtain its motion characteristics; The static obstacle features are obtained by weighted fusion of the image features and the distance features. The dynamic obstacle features are obtained by weighted fusion of the image features, distance features, and motion features.

8. A computer device, characterized in that, The system includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of an automatic parking method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of an automatic parking method as described in any one of claims 1 to 6.

Images