A method of assisting with reversing a vehicle

Abstract

This invention discloses a method for assisting in reversing, comprising: identifying the complete outline of a parking space and determining whether there are obstacles within the outline; if obstacles exist, waiting for the obstacles to be removed or selecting a new parking space; otherwise, using the vehicle's built-in reversing camera to collect video stream information of the ground behind the vehicle and performing adaptive video frame interval sampling; performing semantic segmentation on the extracted frame images to extract the outline information of the parking space outline, pedestrians, and other vehicles in the frame images; using a depth residual network to perform depth estimation to obtain a depth map; analyzing the angular relationship between the vehicle and the parking space outline and the width of the parking space, adjusting the vehicle's pose, calculating and planning the reversing path; obtaining the distances between the vehicle and other vehicles and pedestrians appearing near the reversing location, and analyzing the possibility of reversing based on the estimated reversing path, determining the optimal reversing path based on various possible situations, and adjusting the reversing route in a timely manner.

Description

Technical Field

[0001] This invention relates to a method for assisting in reversing, belonging to the field of path planning technology during vehicle reversing. Background Technology

[0002] With the emergence of deep learning-based optical flow and monocular depth estimation methods, vehicle information in video streams can be quickly identified and detected. A monocular vision-based method for extracting vehicle trajectories in traffic scenes utilizes video streams captured by a fixed panoramic camera and the coordinate information of multiple feature points in the scene to extract vehicle speed and trajectory information. Experiments show that this method has advantages such as fewer constraints, convenient deployment, and accurate recovery, while also ensuring a certain degree of real-time performance, thus possessing practical engineering significance.

[0003] Video analytics-based detection and target recognition technologies have evolved from traditional functions to deep learning. These traditional and intuitive functions achieved good results at this stage. However, due to factors such as the diversity of targets and changes in angle, traditional feature-based target detection has encountered insurmountable bottlenecks. In recent years, deep learning technology has significantly improved the efficiency of detecting and recognizing targets in multiple states. Autonomous driving requires the identification and evaluation of numerous targets, including vehicles, pedestrians, road signs, and reversing assistance, making target recognition environments complex and difficult to implement. From the perspective of pedestrian and vehicle recognition, deep learning-based methods are significantly superior to traditional methods. This is because they can learn to extract features from image data, resulting in better stability and accuracy.

[0004] Currently, the methods for obtaining rear road information when reversing generally involve using lidar, ultrasonic radar, and binocular cameras to establish a measurement system. First, the system acquires road images located in the driver's blind spot behind the vehicle and converts them into 3D point cloud data. Then, it calculates the distance to obstacles using methods such as triangulation, provides processing data, identifies target information through camera target recognition, calculates the distance between the vehicle and these targets, and obtains information about the entire road.

[0005] Some patents and papers involve methods for analyzing the road conditions behind the vehicle to assist in reversing. Some papers use radar image design based on information fusion. For example, "Design and Implementation of Reversing Radar Image System Based on Information Fusion" uses ultrasonic reversing radar and binocular cameras to calculate the distance between the vehicle and obstacles behind it. However, it only calculates the distance to obstacles and cannot analyze path information. Furthermore, the calculation of parking space information is computationally intensive, complex, and time-consuming. Other papers use optical flow detection to extract obstacle information in the parking space. For example, Qiu Xiaogang's paper "Intelligent Reversing Assist System Based on Machine Vision" analyzes distance and detects parking lines using the Hough line detection method to mark the parking position. This method has a narrow scope of application and can only achieve good results in specific situations. It has low accuracy in identifying existing obstacles and poor analysis results. When there is a vehicle next to the parking space, it will also be judged as an obstacle, thus stopping the parking. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method for assisting reversing, which solves the problems of large amount of calculation, high complexity, long time consumption and low accuracy of obstacle recognition in existing assisted reversing technologies, and enhances the safety, accuracy and convenience of the reversing process.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0008] A method for assisting in reversing includes the following steps:

[0009] Step 1: During the process of reversing the vehicle and parking it in the parking space, identify whether there is a complete outline of the parking space on the ground behind the vehicle from the moment the vehicle starts reversing. When a complete outline of the parking space is identified, determine whether there are any obstacles within the identified outline of the parking space.

[0010] Step 2: If there is an obstacle, wait for the obstacle to be removed or select a new parking space and return to Step 1; otherwise, proceed to Step 3.

[0011] Step 3: Use the vehicle's built-in reversing camera to collect video stream information of the ground behind the vehicle, and perform adaptive video frame interval sampling on the video stream information.

[0012] Step 4: Perform semantic segmentation on the frame images extracted in Step 3 to extract the outlines of parking spaces, pedestrians, and other vehicles in the frame images.

[0013] Step 5: For the semantically segmented frame image, use a deep residual network to perform depth estimation to obtain the depth image of the ground behind the vehicle.

[0014] Step 6: Based on the depth image obtained in Step 5, analyze the angular relationship between the vehicle and the outline of the parking space, as well as the width of the parking space, adjust the vehicle's pose, and calculate and plan the reversing path.

[0015] Step 7: Based on the depth image, obtain the distances of other vehicles and pedestrians near the reversing location to the vehicle itself. Based on the reversing path estimated in Step 6, analyze the possibility of reversing, determine the optimal reversing path based on various possible scenarios, and adjust the reversing route in a timely manner so that the car can be safely parked in the parking space.

[0016] As a preferred embodiment of the present invention, in the adaptive video frame interval sampling in step 3, the current sampling interval time is adaptively adjusted according to the current vehicle speed and the maximum detection distance of the deep residual network. The specific sampling interval time is as follows:

[0017]

[0018] Where V represents the current vehicle speed, and T(V) represents the sampling interval time at vehicle speed V.

[0019] In a preferred embodiment of the present invention, in step 4, the semantic segmentation network detectron2 is used to perform semantic segmentation operations on the frame images extracted in step 3.

[0020] In a preferred embodiment of the present invention, in step 5, a Laplace pyramid structure is used in the decoder of the deep residual network, the deep residual is defined by the decomposition of the pyramid, and the activation function of the deep residual network is the Softplus activation function, i.e.:

[0021] Softplus(x) = ln(1 + exp(x))

[0022] Where x represents the input to the Softplus activation function;

[0023] The conversion relationship between the vehicle's built-in reversing camera and the actual distance is calculated to obtain a depth image of the ground behind the vehicle. The calculation formula is as follows:

[0024]

[0025] Among them, (x c ,y c ,z c P represents the pixel point P within the depth map obtained from depth estimation. c Coordinates, (x, y, z) are coordinates in the world coordinate system, k and l are camera intrinsic parameters, f is the camera extrinsic parameter, c x c is the offset of the x-coordinate in the pixel coordinate system. y This represents the offset of the y-coordinate in the pixel coordinate system.

[0026] As a preferred embodiment of the present invention, the specific process of step 6 is as follows:

[0027] Step 6.1: Based on the parking space outline information in the depth image obtained in Step 5, determine the width w of the parking space, i.e.:

[0028] w = y1 - y2

[0029] Define the two longer contour lines in the parking space outline as the left and right contour lines, with y1 being the ordinate value of the left contour line and y2 being the ordinate value of the right contour line.

[0030] Step 6.2, calculate the angle between the vehicle itself and the right profile line:

[0031]

[0032] Step 6.3: Based on the angle θ obtained in Step 6.2, obtain the vehicle's own pose information, i.e., the vehicle's position relative to the parking space, and calculate the optimal reversing route, specifically:

[0033] Establish a two-dimensional coordinate system with the vehicle's built-in reversing camera as the origin and the vehicle's front orientation as the positive Y-axis. Then, the coordinates of the midpoint A of the vehicle body in this two-dimensional coordinate system are (x...). a ,y a ,α a The coordinates of the midpoint B of the parking space in the two-dimensional coordinate system are (x, y). b ,y b ,α b ), where α a α has the same magnitude as angle θ but opposite direction; b The size is The direction is the direction the car is facing; extend the shorter centerline of the parking space and the longer centerline of the vehicle, respectively, find the intersection point D of the two extended lines, and obtain the coordinates (x, y) of the intersection point D. d ,y d Using the distance formula, the distance h between points B and D is calculated, which is the turning radius R of the car reversing into the parking space. A circle is drawn based on this turning radius, and lines AD and BD are both tangent to this circle. The point of tangency between the circle and line AD is the turning point C. The coordinates of turning point C are then calculated. The arc is then determined based on the turning radius R and the coordinates of points B and C. The length of this distance is the turning distance;

[0034]

[0035]

[0036] That is, the vehicle first reverses to point C, and then, based on the turning radius and the arc... Measure the length and reverse along the arc to point B, thus completing the reversing and parking operation in the parking space.

[0037] As a preferred embodiment of the present invention, the specific process of step 7 is as follows:

[0038] Step 7.1: If other vehicles or pedestrians appear near the reversing location, determine the distance between the other vehicles and the vehicle itself, the distance between the pedestrians and the vehicle itself, the speed of the other vehicles, and the speed of the pedestrians based on the current frame image and the previous frame image.

[0039] Step 7.2: If other vehicles appear behind your vehicle in the current frame image, calculate the distance D between the other vehicles and your vehicle based on the current frame image. hm The distance D between the vehicle and the parking space d When D hm Less than D d When it is determined that the parking space cannot be reached, stop reversing, select a new parking space, and return to step 1. hm Greater than D d If it is determined that reversing is permissible, then during the reversing process, the following judgment formula is used to further determine whether reversing is still possible:

[0040]

[0041] H = Kh

[0042]

[0043] Among them, D s v1 is the distance between the vehicle and other vehicles during the reversing process, v2 is the reversing speed planned in step 6, v2 is the speed of other vehicles, H is the safe distance, K is the reversing distance planned in step 6, h is the error, and T is the reversing time planned in step 6.

[0044] When H is greater than D s If it is determined that other vehicles are affecting the reversing process, stop reversing and continue reversing after the other vehicles have left. When H is less than D... s If other vehicles do not affect the reversing process, proceed to step 7.3 if a pedestrian appears behind the vehicle during the reversing process, and continue to step 7.2 after step 7.3 is completed.

[0045] Step 7.3: If a pedestrian appears behind the vehicle in the current frame image, record the pedestrian's coordinates (x, y). o ,y o ,z o ), calculate the distance between the pedestrian and your vehicle:

[0046]

[0047] Determine whether reversing is permissible based on the judgment formula:

[0048] F = K + h

[0049] When L is greater than F, it is determined that the pedestrian does not affect the reversing motion. When L is less than F, it is determined that the pedestrian affects the reversing motion, so the reversing motion is stopped and continues after the pedestrian leaves.

[0050] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0051] 1. This invention utilizes deep residual networks for depth estimation and 3D image reconstruction to obtain real-time information about the surrounding environment during reversing (including obstacles, vehicles, pedestrians, etc. on the reversing route), and plans the optimal reversing path based on the real-time situation.

[0052] 2. Compared with other reversing methods, this invention converts two-dimensional images into three-dimensional images. Based on the three-dimensional images, more accurate depth and distance information can be obtained, which improves the accuracy of distance estimation and the safety of reversing.

[0053] 3. This invention uses a monocular camera. Compared with a binocular camera, a monocular camera has higher timeliness and can reflect road information in a shorter time. Monocular cameras are also easy to install and have lower costs. Attached Figure Description

[0054] Figure 1 This is a flowchart of a method for assisting in reversing according to the present invention;

[0055] Figure 2 This is a diagram of the deep residual network structure used in this invention;

[0056] Figure 3 This is a diagram showing the effect of the depth estimation algorithm used in this invention;

[0057] Figure 4 This is the reversing parking planning diagram of the present invention. Detailed Implementation

[0058] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0059] This invention proposes an assisted reversing method that fully utilizes the latent characteristics of well-encoded feature targets for monocular depth estimation. This invention introduces the Laplacian pyramid model, using images in a pyramid structure as network input. The pyramid structure input enables the network to extract features at multiple scales. The features are input into different video streams to decode depth residual operations, thereby obtaining road information (including the size, coordinates, distance, speed, etc. of the vehicle and surrounding objects) during the reversing process, which is used to rationally plan the path during the reversing process.

[0060] Acquiring depth images of the surrounding environment of a reversing vehicle is the key technology of this invention. Depth estimation of the acquired image information based on deep learning is the core of this invention. This invention uses a depth residual network based on a Laplacian pyramid structure to estimate depth and make a judgment on road conditions.

[0061] like Figure 1 The diagram shows a flowchart of a method for assisting in reversing according to the present invention. The specific steps are as follows:

[0062] Step 1: During the process of reversing into the parking space in the parking lot, identify whether there is an outline on the ground behind the vehicle. When a complete outline of a parking space is identified, detect whether there are any obstacles within the outline.

[0063] Step Two: If there are obstacles, wait for them to be removed or select a new parking space, and then identify and detect them again. When there are no obstacles within the outline, i.e., the selected parking space is free of obstacles, it is selected as the reversing location, and you can reverse to the next step.

[0064] Step 3: Use the vehicle's built-in reversing camera to collect video stream data of the ground behind the vehicle, and perform adaptive video frame interval sampling on the video stream data.

[0065] For the current vehicle speed, it's crucial to ensure effective frame acquisition within each vehicle interval to guarantee valid input for the network model, prevent congestion-induced delays, and improve performance. The default maximum detection range for the deep residual network is 40-50 meters. With a current vehicle speed of V, the default interval sampling time formula is:

[0066]

[0067] Step 4: Use a semantic segmentation network to segment the extracted frame images and extract the parking line at the reversing location, as well as the outline information of pedestrians and vehicles near the reversing location.

[0068] Using a semantic segmentation network: The semantic segmentation network of detectron2 is used to perform segmentation operations on the extracted frame images, extracting the outline of the parking space at the reversing location, as well as the outline information of pedestrians and other vehicles near the reversing location.

[0069] Step 5: Perform depth estimation on the semantically segmented frame images to obtain the subsequent depth image.

[0070] Step 501: Construct a deep residual convolutional network using the Laplacian pyramid structure; Figure 2 This is a diagram of the deep residual network structure used in this invention; Figure 3 This is a diagram showing the effect of the depth estimation algorithm used in this invention.

[0071] A Laplacian pyramid structure is used in the decoder of the deep residual convolutional network, utilizing the pyramid decomposition to define the depth residual. At the decoder end, a pixel reassembly upsampling module is designed. This module uses pixel reassembly to fuse global features and pyramid features, improving feature resolution. Specifically, the pyramid module extracts feature information at various scales, while pixel reassembly improves feature resolution and preserves all feature information. Experiments show that this method enhances feature information, improves the accuracy of depth estimation, and can accurately estimate the depth boundaries of targets within road conditions and provide a comprehensive understanding of the global context.

[0072] Step 502, Activation function modification

[0073] Activation functions are used to transform nonlinear combinations and hidden layers into linear ones. In this invention, the Softplus activation function is selected, and its function definition is as follows:

[0074] Softplus(x) = ln(1 + exp(x))

[0075] Both Softplus and ReLU are closer to the activation model of brain neurons. Softplus was chosen as the activation function because the data processed by the Softplus function is smoother than that of the original ReLU function, while also maintaining the unidirectional inhibitory property of ReLU.

[0076] Step 503: Calculate the conversion relationship between camera and actual distance.

[0077] Because different cameras are used, the distance that depth estimation can reflect also varies. The formula for the actual distance corresponding to pixels in a photo is:

[0078]

[0079] Among them, (x c ,y c ,z c P represents the pixel point P within the depth map obtained from depth estimation.c Coordinates, (x, y, z) are coordinates in the world coordinate system, k and l are camera intrinsic parameters, f is the camera extrinsic parameter, c x c is the offset of the x-coordinate in the pixel coordinate system. y This represents the offset of the y-coordinate in the pixel coordinate system.

[0080] Step Six: Analyze the angular relationship between the vehicle and the parking line and the width of the empty space, adjust the vehicle's position, and calculate and plan the reversing path.

[0081] Step 601: Determine the width w of the parking space based on the parking space outline information in the depth map. The formula is:

[0082] w = y1 - y2

[0083] Define the two longer contour lines in the parking space outline as the left and right contour lines, with y1 being the ordinate value of the left contour line and y2 being the ordinate value of the right contour line.

[0084] Step 602: Calculate the angle between the vehicle body and the parking lines on both sides of the reversing position:

[0085]

[0086] Where arccos is the inverse function of cos, and θ is the angle between the vehicle and the right profile line.

[0087] Step 603: Based on θ obtained in step 602, the vehicle's own pose information (i.e., the vehicle's position relative to the parking space) can be obtained, and the optimal reversing route can be calculated as follows: Figure 4 As shown. Establish a two-dimensional coordinate system with the camera as the origin and the car's heading as the positive Y-axis. The coordinates of the midpoint A of the car body in the two-dimensional plane are (x, y). a ,y a ,α a And the coordinates (x, y) of the midpoint B of the parking space in the two-dimensional plane. b ,y b ,α b ), where α a Same magnitude as θ, but opposite direction, α b The magnitude is π / 2 - θ, representing the direction the car is facing; find the intersection point D of the two extensions along the direction the car is facing, and then calculate the coordinates of the intersection point D(x). d ,y d The distance h between points B and D is calculated using the distance formula, and h is the turning radius R of the car reversing into the parking space. The coordinates C of the turning point are obtained by using the turning radius R, the fact that lines AD and BD are all tangent to the circle containing radius R, and the radius R of the circle and the coordinates of points B and C to determine the arc. The length of this distance is the turning distance;

[0088] in,

[0089]

[0090]

[0091] The vehicle will first reverse to point C, and then, based on the turning radius and the arc... Measure the length and reverse along the arc to point B, thus completing the reversing and parking operation in the parking space.

[0092] Step 7: Based on the depth image behind, obtain the distances of vehicles, pedestrians, etc. that appear near the reversing location, and analyze the possibility of reversing based on the reversing path estimated in Step 6. Analyze the optimal reversing path based on various possible situations, and adjust the reversing route in a timely manner so that the car can safely reverse to the planned location.

[0093] Step 701: If other vehicles or pedestrians appear near the reversing location, determine the distance and position information of the vehicles and pedestrians based on two consecutive depth images. The acquisition interval is 10ms. Estimate the distance to other obstacles and the required speed every 10ms and replace the previous speed and distance information.

[0094] Step 702: During the reversing process, determine the speed and distance of the objects behind, and find that the distance to the approaching vehicle is D. hm The distance from the destination (parking space) to the reverse driving destination is D. d When D hm Less than D d When the system determines that the destination cannot be reached, it stops reversing and replans the route. hm Greater than D d When the system determines that it can continue reversing, if an obstacle or pedestrian is detected behind, the processor executes a stop and wait command, waiting for the next instruction to be issued.

[0095] The judgment formula is as follows:

[0096]

[0097] H = Kh

[0098]

[0099] Where v1 is the originally planned reversing speed, v2 is the speed of vehicles, pedestrians, obstacles, etc. on the road behind; T is the planned reversing time, K is the originally planned reversing distance, and D is the reversing distance. s The distance between the vehicle and the following vehicle within time T is the distance of the vehicle itself; h is the error, and the value of h is usually in the range of 3 to 5 m; H is the safe following distance.

[0100] When H is greater than D s When it is determined that vehicles on the road behind are affecting the reversing process, reversing is not allowed, and H is less than D. s If it is determined that vehicles on the road behind do not obstruct the reversing process, then reversing is permitted.

[0101] Step 703: If a pedestrian appears behind the vehicle while it is reversing, record the pedestrian's coordinates (x, y, y). o ,y o ,z o Using the formula:

[0102]

[0103] Where L is the distance between the pedestrian and the vehicle; based on the decision function F:

[0104] F = K + h

[0105] When L is greater than F, it is determined that the road behind is safe for reversing. When L is less than F, it is determined that there are pedestrians behind and reversing is not allowed, so the reversing behavior is stopped immediately.

[0106] The above embodiments are merely illustrative of the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solutions based on the technical concept proposed in this invention shall fall within the scope of protection of this invention.

Claims

1. A method for assisting in reversing, characterized in that, Includes the following steps: Step 1: During the process of reversing the vehicle and parking it in the parking space, identify whether there is a complete outline of the parking space on the ground behind the vehicle from the moment the vehicle starts reversing. When a complete outline of the parking space is identified, determine whether there are any obstacles within the identified outline of the parking space. Step 2: If there is an obstacle, wait for the obstacle to be removed or select a new parking space and return to Step 1; otherwise, proceed to Step 3. Step 3: Use the vehicle's built-in reversing camera to collect video stream information of the ground behind the vehicle, and perform adaptive video frame interval sampling on the video stream information. Step 4: Perform semantic segmentation on the frame images extracted in Step 3 to extract the outlines of parking spaces, pedestrians, and other vehicles in the frame images. Step 5: For the semantically segmented frame image, use a deep residual network to perform depth estimation to obtain the depth image of the ground behind the vehicle. Step 6: Based on the depth image obtained in Step 5, analyze the angular relationship between the vehicle and the outline of the parking space, as well as the width of the parking space, adjust the vehicle's pose, and calculate and plan the reversing path. Step 7: Based on the depth image, obtain the distances between your vehicle and other vehicles and pedestrians near the reversing location. Analyze the reversing path estimated in Step 6 to determine the probability of successful reversing. Based on various possible scenarios, determine the optimal reversing path and adjust the reversing route accordingly to ensure the car safely parks in the parking space. The specific process is as follows: Step 7.1: If other vehicles or pedestrians appear near the reversing location, determine the distance between the other vehicles and the vehicle itself, the distance between the pedestrians and the vehicle itself, the speed of the other vehicles, and the speed of the pedestrians based on the current frame image and the previous frame image. Step 7.2: If other vehicles appear behind your vehicle in the current frame image, calculate the distance D between the other vehicles and your vehicle based on the current frame image. hm The distance D between the vehicle and the parking space d When D hm Less than D d When it is determined that the parking space cannot be reached, stop reversing, select a new parking space, and return to step 1. hm Greater than D d If it is determined that reversing is permissible, then during the reversing process, the following judgment formula is used to further determine whether reversing is still possible: H = Kh Among them, D s v1 is the distance between the vehicle and other vehicles during the reversing process, v2 is the reversing speed planned in step 6, v2 is the speed of other vehicles, H is the safe distance, K is the reversing distance planned in step 6, h is the error, and T is the reversing time planned in step 6. When H is greater than D s If it is determined that other vehicles are affecting the reversing process, stop reversing and continue reversing after the other vehicles have left. When H is less than D... s If other vehicles do not affect the reversing process, proceed to step 7.3 if a pedestrian appears behind the vehicle during the reversing process, and continue to step 7.2 after step 7.3 is completed. Step 7.3: If a pedestrian appears behind the vehicle in the current frame image, record the pedestrian's coordinates (x, y). o ,y o ,z o ), calculate the distance between the pedestrian and the vehicle: Determine whether reversing is permissible based on the judgment formula: F = K + h When L is greater than F, it is determined that the pedestrian does not affect the reversing motion. When L is less than F, it is determined that the pedestrian affects the reversing motion, so the reversing motion is stopped and continues after the pedestrian leaves.

2. The method for assisting in reversing according to claim 1, characterized in that, In step 3, the adaptive video frame interval sampling is performed by adaptively adjusting the current sampling interval time based on the current vehicle speed and the maximum detection distance of the deep residual network. The specific sampling interval time is as follows: Where V represents the current vehicle speed, and T(V) represents the sampling interval time at vehicle speed V.

3. The method for assisting in reversing according to claim 1, characterized in that, In step 4, the semantic segmentation network detectron2 is used to perform semantic segmentation on the frame images extracted in step 3.

4. The method for assisting in reversing according to claim 1, characterized in that, In step 5, a Laplace pyramid structure is used in the decoder of the deep residual network. The deep residual is defined using the decomposition of the pyramid, and the activation function of the deep residual network is the Softplus activation function, i.e.: Softplus(x) = ln(1 + exp(x)) Where x represents the input to the Softplus activation function; The conversion relationship between the vehicle's built-in reversing camera and the actual distance is calculated to obtain a depth image of the ground behind the vehicle. The calculation formula is as follows: Among them, (x c ,y c ,z c P represents the pixel point P within the depth map obtained from depth estimation. c Coordinates, (x, y, z) are coordinates in the world coordinate system, k and l are camera intrinsic parameters, f is the camera extrinsic parameter, c x c is the offset of the x-coordinate in the pixel coordinate system. y This represents the offset of the y-coordinate in the pixel coordinate system.

5. The method for assisting in reversing according to claim 1, characterized in that, The specific process of step 6 is as follows: Step 6.1: Based on the parking space outline information in the depth image obtained in Step 5, determine the width w of the parking space, i.e.: w = y1 - y2 Define the two longer contour lines in the parking space outline as the left and right contour lines, with y1 being the ordinate value of the left contour line and y2 being the ordinate value of the right contour line. Step 6.2, calculate the angle between the vehicle itself and the right profile line: Step 6.3: Based on the angle θ obtained in Step 6.2, obtain the vehicle's own pose information, i.e., the vehicle's position relative to the parking space, and calculate the optimal reversing route, specifically: Establish a two-dimensional coordinate system with the vehicle's built-in reversing camera as the origin and the vehicle's front orientation as the positive Y-axis. Then, the coordinates of the midpoint A of the vehicle body in this two-dimensional coordinate system are (x...). a ,y a ,α a The coordinates of the midpoint B of the parking space in the two-dimensional coordinate system are (x, y). b ,y b ,α b ), where α a α has the same magnitude as angle θ but opposite direction; b The size is The direction is the direction the car is facing; extend the shorter centerline of the parking space and the longer centerline of the vehicle, respectively, find the intersection point D of the two extended lines, and obtain the coordinates (x, y) of the intersection point D. d ,y d Using the distance formula, the distance h between points B and D is calculated, which is the turning radius R of the car reversing into the parking space. A circle is drawn based on this turning radius, and lines AD and BD are both tangent to this circle. The point of tangency between the circle and line AD is the turning point C. The coordinates of turning point C are then calculated. The arc is then determined based on the turning radius R and the coordinates of points B and C. The length of this distance is the turning distance; That is, the vehicle first reverses to point C, and then, based on the turning radius and the arc... Measure the length and reverse along the arc to point B, thus completing the reversing and parking operation in the parking space.

Images