Parking scene wall reconstruction method based on TSDF and processor

Abstract

The invention relates to the technical field of parking scene establishment, in particular to a parking scene wall reconstruction method based on TSDF and a processor, and the reconstruction method comprises the following steps: sensing and extracting wall scatter points; searching a sequence of scattered points on the wall surface; and performing linear interpolation between adjacent points according to the sequence of the scattered points of the wall surface, and filtering and outputting the edge of the wall surface. The method solves the problems that when an existing TSDF algorithm processes wall reconstruction, high-density wall scatter points cannot be obtained, the calculation amount is sharply increased due to high-density sensing data, optimization of composition efficiency is affected, and an exponential optimization strategy is difficult to achieve.

Description

Technical Field

[0001] This invention relates to the field of parking scene creation technology, and in particular to a parking scene wall reconstruction method and processor based on TSDF. Background Technology

[0002] In recent years, with the advancement of automatic parking technology and the reduction in costs, the pre-installation penetration rate of automatic parking in new cars in China has continued to rise. Since the success rate of automatic parking largely depends on the success rate and accuracy of parking space recognition, parking space recognition technology also needs to continue to improve accordingly.

[0003] In parking scenarios using parking space recognition technology, walls are a crucial element requiring high-precision reconstruction. However, due to the diverse types of walls, including curved walls, straight single walls, and corner walls, existing structured fitting algorithms cannot handle complex walls and cannot meet the intermediate requirements for all wall types. Therefore, 3D reconstruction technology is needed to address this issue.

[0004] Among existing 3D reconstruction technologies, TSDF is widely known. However, the perceptual data generally used by the TSDF algorithm is mostly high-density and low-noise images combined with depth maps or high-density laser point cloud maps. But the initial extraction of wall scatter points that meet these characteristics requires maintaining a high density, which poses a high challenge to the perceptual algorithm. At the same time, this high-density perceptual data leads to a sharp increase in the amount of computation in the subsequent mapping process, which in turn affects the optimization of mapping efficiency and makes it difficult to achieve exponential optimization strategies. Summary of the Invention

[0005] The present invention provides a method and processor for wall reconstruction in parking scenes based on TSDF. It mainly solves the problems that existing TSDF algorithms cannot obtain high-density wall scatter points when processing wall reconstruction, and the computational load caused by high-density sensing data increases sharply, which affects the optimization of mapping efficiency and makes it difficult to achieve exponential optimization strategies.

[0006] This invention proposes a method for reconstructing walls in a parking scene based on TSDF, comprising the following steps:

[0007] Sensory extraction of scattered points on the wall surface;

[0008] The order in which to find scattered points on the wall;

[0009] Linear interpolation is performed between adjacent points according to the order of scattered points on the wall, and the wall edge is output after filtering.

[0010] Preferably, the order of finding scattered points on the wall includes:

[0011] In the parking scene coordinate system, determine whether all the scattered points on the wall are oriented along the x-axis or y-axis, and obtain the starting point of all the scattered points on the wall based on the x-axis or y-axis orientation.

[0012] The order in which the scattered points on the wall are connected sequentially from the starting point to the adjacent scattered points on the wall is used to find the scattered points on the wall.

[0013] Preferably, determining whether all the wall points are along the x-axis or y-axis in the parking scene coordinate system, and obtaining the starting point of all the current wall points based on the x-axis or y-axis direction, includes:

[0014] In the parking scene coordinate system, determine the distance between the maximum and minimum x-axis values ​​of all the scattered points on the wall, and the distance between the maximum and minimum y-axis values ​​of all the scattered points on the wall;

[0015] Compare the calculated x-axis distance with the y-axis distance. If the x-axis distance is greater than the y-axis distance, then all the scattered points on the wall are determined to be along the x-axis; otherwise, they are determined to be along the y-axis.

[0016] If the direction is along the x-axis, the starting point is the minimum value of the x-axis; if the direction is along the y-axis, the starting point is the minimum value of the y-axis.

[0017] Preferably, starting from the minimum value of the x-axis, or starting from the minimum value of the y-axis, includes:

[0018] Find the minimum point on the x-axis or y-axis;

[0019] Determine whether the current minimum point has any adjacent points. If so, use the adjacent points as the minimum point and determine again until any minimum point has no adjacent points. Then, use the current minimum point as the starting point of all the scattered points on the wall.

[0020] Preferably, determining whether the current minimum point has adjacent points includes:

[0021] Determine whether the current minimum point has adjacent points in each of its eight neighborhoods. If there are at least two adjacent points, then determine each of the at least two adjacent points one by one.

[0022] Preferably, the sensing and extraction of scattered points on the wall surface includes:

[0023] The sensor fusion outputs multiple scattered points, which are then subjected to DBSCAN clustering.

[0024] Divide all scattered points into a grid;

[0025] Multiple rays are formed outward from the observation point. It is determined whether any ray passes through any of the scattered points. If so, the current scattered point is defined as the wall scattered point.

[0026] Preferably, determining whether any of the rays passes through any of the scattered points, and defining the current scattered point as the wall scattered point if so, includes:

[0027] Determine whether the current ray passes through a scatter point and whether the number of scatter points is greater than two. If so, define the scatter point closest to the observation point as the wall scatter point; otherwise, define the unique scatter point as the wall scatter point.

[0028] Preferably, the filtered output scene wall includes:

[0029] Define a signal sequence x[n] containing N points, where n represents the index of the sample [0, N-1].

[0030] A window of fixed length M is formed by sliding across the signal sequence, and the average value of the samples within the window is calculated and filtered.

[0031]

[0032] Where y[n] is the filtered output signal; x[n] is the input signal; N is the size of the filtering window; and k is the sample index within the window.

[0033] Preferably, the output wall edge is used as the edge between the wall and the ground, and the wall is extended upward along the wall edge to form a wall, and then rendered to form a scene wall.

[0034] The present invention also proposes a processor, the processor including a control chip, the control chip being used to implement the aforementioned method for reconstructing the wall in a parking scene.

[0035] As can be seen from the above, the following beneficial effects can be obtained by applying the technical solution provided by the present invention:

[0036] First, the parking scene wall reconstruction method proposed in this invention can use the TSDF algorithm to search the order of the scattered points on the wall and determine whether the current scattered points on the wall are in the x-axis or y-axis direction based on the search results, thereby improving the accuracy of the algorithm detection and reducing the amount of computation.

[0037] Second, in the parking scene wall reconstruction method proposed in this invention, the scattered points extracted by perception are relatively sparse. Therefore, by emitting rays along the detection points, the scattered points of the wall are determined, so that the output scattered points of the wall meet the input requirements of TSDF. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a flowchart illustrating the specific steps for finding scattered points on the wall surface in an embodiment of the present invention. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0041] Existing TSDF algorithms cannot acquire high-density wall scatter points when processing wall reconstruction, and the computational load caused by high-density sensing data increases dramatically, which in turn affects the optimization of mapping efficiency and makes it difficult to achieve exponential optimization strategies.

[0042] like Figure 1 As shown, to solve the above problems, this embodiment proposes a TSDF-based method for reconstructing walls in parking scenarios, including the following steps:

[0043] S1, perceive and extract scattered points on the wall;

[0044] S2, find the order of scattered points on the wall;

[0045] S3 performs linear interpolation between adjacent points according to the order of the scattered points on the wall, and then filters and outputs the wall edge.

[0046] Step S1 includes:

[0047] S11, Perception fusion outputs multiple scattered points for DBSCAN clustering;

[0048] S12, divide all scattered points onto the grid;

[0049] S13: Multiple rays are formed outward from the observation point. It is determined whether any ray passes through any scatter point. If so, the current scatter point is defined as the wall scatter point.

[0050] Preferably, in step S12, each grid cell has one and only one scatter point.

[0051] More specifically, step S13 includes: determining whether the current ray passes through a scatter point and whether the number of scatter points is greater than two. If so, the scatter point closest to the observation point is defined as the wall scatter point; otherwise, a unique scatter point is defined as the wall scatter point.

[0052] Preferably, in this embodiment, the rays emitted by the observation point are not obtained from a single observation, but are determined by adjusting the position of the observation point multiple times and repeatedly emitting rays and obtaining scattered points.

[0053] Preferably, in this embodiment, the radiation density emitted by the observation point is relatively high, thus making the trend between the detected wall scattered points clear. Specifically, the user can automatically judge and adjust the radiation density. Alternatively, in this embodiment, the radiation density is stable. The ray is used to determine which single cell contains scattered points among all the grids that the current ray passes through, and the scattered points in the corresponding grid are taken as the scattered points that the ray passes through. The distance to all the scattered points that the ray passes through is then judged. For example, under the calibration ray, the ray passes through grids numbered 1 to 7, and there are scattered points in grids 1 to 3. Therefore, the distance to the above three scattered points is judged to determine that scattered point 3 should be a wall scattered point.

[0054] More specifically, step S2 includes:

[0055] S21, in the parking scene coordinate system, determine whether all wall scattered points are oriented along the x-axis or y-axis, and obtain the starting point of all current wall scattered points based on the x-axis or y-axis orientation;

[0056] S22, connect the scattered points on the adjacent walls from the starting point to find the order of the scattered points on the walls.

[0057] Step S21 includes:

[0058] S211, in the parking scene coordinate system, determine the distance between the maximum and minimum x-axis values ​​of all wall scattered points, and the distance between the maximum and minimum y-axis values ​​of all wall scattered points;

[0059] S212, compare the calculated x-axis distance with the y-axis distance. If the x-axis distance is greater than the y-axis distance, then all wall points are determined to be along the x-axis; otherwise, they are determined to be along the y-axis.

[0060] S213, if the direction is x-axis, then the minimum value of x-axis is the starting point; if the direction is y-axis, then the minimum value of y-axis is the starting point.

[0061] In step S213, starting from the minimum value of the x-axis or the minimum value of the y-axis, the following are included:

[0062] Find the minimum point on the x-axis or y-axis;

[0063] Determine if the current minimum point has any adjacent points. If so, use the adjacent points as the minimum point and check again until no minimum point has any adjacent points. Then, use the current minimum point as the starting point for all wall scattered points.

[0064] The process of determining whether the current minimum point has adjacent points includes:

[0065] Determine whether the current minimum point has adjacent points in each of its eight neighborhoods. If there are at least two adjacent points, then check each of the at least two adjacent points one by one.

[0066] In this embodiment, the distance between the maximum and minimum x-axis values ​​in step S211 is actually obtained by acquiring the x-values ​​of the two endpoints of the x-axis that are furthest apart on the wall surface, and calculating Δx. Similarly, after calculating Δy, the magnitude relationship between Δx and Δy is compared.

[0067] In this embodiment, the acquisition of adjacent points at the starting point can be achieved using point grabbing technology, or it can be determined by whether there are scattered points in the adjacent grids of the grid where the current scattered point is located on the aforementioned grid map. The adjacent grids can be set with a spacing geometry to ensure full coverage of the grabbing.

[0068] More specifically, step S3 includes:

[0069] S31, Define a signal sequence x[n] containing N points, where n represents the index of the sample [0, N-1];

[0070] S32, slide a window of fixed length M across the signal sequence, calculate the average value of the samples within the window, and perform filtering;

[0071]

[0072] Where y[n] is the filtered output signal; x[n] is the input signal; N is the size of the filtering window; and k is the sample index within the window.

[0073] In this embodiment, the larger N is, the better the filtering effect is for dense scattered points, but the worse the effect is for sparse scattered points; the smaller N is, the better the effect is. Therefore, in this embodiment, linear interpolation must be performed before filtering.

[0074] In this embodiment, based on the obtained wall surface scatter point order, linear interpolation is performed between adjacent points, and the interpolation density is controlled by a difference replacement step.

[0075] More specifically, it also includes the following steps:

[0076] S4 takes the output wall edge as the edge between the wall and the ground, extends upward along the wall edge to form the wall, and renders it to form the scene wall.

[0077] It should be emphasized that the TSDF-based wall reconstruction method for parking scenes described in this embodiment addresses the issue that TSDF's scattered points are sparse when applied to wall reconstruction, and each ray cannot guarantee an outer surface endpoint. This is addressed by using the method of checking if the ray path of the observation point contains an outer surface endpoint. Furthermore, to address the issue of larger errors at distant pixels in TSDF, weights are removed since there is no difference in observation errors between different points. Finally, regarding the problem of only constructing the outer surface endpoints within the current camera's view frustum, since the current scattered points are not observed in a single camera observation but rather accumulated from multiple observations, the current verification demonstrates that this is solved by selecting the closest observation point using a kd-tree and constructing a ray.

[0078] In summary, the TSDF-based wall reconstruction method for parking scenarios proposed in this embodiment can be optimized using different algorithms according to the actual parking environment, while only using the mature TSDF framework. This method is relatively stable and can ensure the authenticity of the output effect.

[0079] The embodiments described above do not constitute a limitation on the scope of protection of this technical solution. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the above embodiments should be included within the scope of protection of this technical solution.

Claims

1. A method for reconstructing walls in a parking scene based on TSDF, characterized in that, Includes the following steps: Sensory extraction of scattered points on the wall surface; The order in which to find scattered points on the wall; Linear interpolation is performed between adjacent points according to the order of scattered points on the wall, and the wall edge is output after filtering.

2. The method for reconstructing parking scene walls based on TSDF according to claim 1, characterized in that, The order of finding scattered points on the wall includes: In the parking scene coordinate system, determine whether all the scattered points on the wall are oriented along the x-axis or y-axis, and obtain the starting point of all the scattered points on the wall based on the x-axis or y-axis orientation. The order in which the scattered points on the wall are connected sequentially from the starting point to the adjacent scattered points on the wall is used to find the scattered points on the wall.

3. The method for reconstructing parking scene walls based on TSDF according to claim 2, characterized in that, The step of determining whether all the scattered points on the wall surface are oriented along the x-axis or y-axis in the parking scene coordinate system, and obtaining the starting point of all the scattered points on the wall surface based on the x-axis or y-axis orientation, includes: In the parking scene coordinate system, determine the distance between the maximum and minimum x-axis values ​​of all the scattered points on the wall, and the distance between the maximum and minimum y-axis values ​​of all the scattered points on the wall; Compare the calculated x-axis distance with the y-axis distance. If the x-axis distance is greater than the y-axis distance, then all the scattered points on the wall are determined to be along the x-axis; otherwise, they are determined to be along the y-axis. If the direction is along the x-axis, the starting point is the minimum value of the x-axis; if the direction is along the y-axis, the starting point is the minimum value of the y-axis.

4. The method for reconstructing parking scene walls based on an improved TSDF according to claim 3, characterized in that, The starting point being the minimum value on the x-axis, or the starting point being the minimum value on the y-axis, includes: Find the minimum point on the x-axis or y-axis; Determine whether the current minimum point has any adjacent points. If so, use the adjacent points as the minimum point and determine again until any minimum point has no adjacent points. Then, use the current minimum point as the starting point of all the scattered points on the wall.

5. The method for reconstructing parking scene walls based on TSDF according to claim 4, characterized in that, Determining whether the current minimum point has adjacent points includes: Determine whether the current minimum point has adjacent points in each of its eight neighborhoods. If there are at least two adjacent points, then determine each of the at least two adjacent points one by one.

6. A method for reconstructing parking scene walls based on TSDF according to any one of 1 to 5, characterized in that, The sensing and extraction of scattered points on the wall surface includes: The sensor fusion outputs multiple scattered points, which are then subjected to DBSCAN clustering. Divide all scattered points into a grid; Multiple rays are formed outward from the observation point. It is determined whether any ray passes through any of the scattered points. If so, the current scattered point is defined as the wall scattered point.

7. The method for reconstructing parking scene walls based on TSDF according to claim 6, characterized in that, The step of determining whether any of the rays passes through any of the scattered points, and defining the current scattered point as the wall scattered point if so, includes: Determine whether the current ray passes through a scatter point and whether the number of scatter points is greater than two. If so, define the scatter point closest to the observation point as the wall scatter point; otherwise, define the unique scatter point as the wall scatter point.

8. The method for reconstructing parking scene walls based on TSDF according to claim 7, characterized in that, The filtered output scene wall includes: Define a signal sequence x[n] containing N points, where n represents the index of the sample. [0，N-1]； A window of fixed length M is formed by sliding across the signal sequence, and the average value of the samples within the window is calculated and filtered. Where y[n] is the filtered output signal; x[n] is the input signal; N is the size of the filtering window; and k is the sample index within the window.

9. The method for reconstructing parking scene walls based on TSDF according to claim 1, characterized in that, Also includes: The output wall edge is used as the edge between the wall and the ground. The wall is extended upward along the wall edge to form the wall surface, and then rendered to form the scene wall surface.

10. A processor, characterized in that: The processor includes a control chip, which is used to implement the parking scene wall reconstruction method according to any one of claims 1 to 9.

Images