Negative obstacle detection method, system, medium and device applied to structured roads

Abstract

The present invention discloses a negative obstacle detection method, system, medium and device applied to a structured road. The method includes the steps of: acquiring the original point cloud information of the environmental scene; constructing a ground model based on the original point cloud information, and extracting ground information through the ground model; taking the extracted ground information as prior information, performing gradient characteristic analysis and geometric distribution statistical analysis on the rasterized adjacent point clouds beam by beam, and taking the analysis results within the sensing range of each beam as candidate negative obstacle grids; summarizing the candidate negative obstacle grids of all beams and performing filtering to output the final negative obstacle detection result. The system, medium and device of the present invention all correspond to the method. The present invention has the advantages of improving the stability and robustness of negative obstacle detection and improving the detection accuracy, etc.

Description

Technical Field

[0001] The present invention mainly relates to the field of driverless technology, and particularly relates to a negative obstacle detection method, system, medium and device applied to structured roads. Background Art

[0002] Negative obstacle perception is the research focus and difficulty in the field of environment perception of driverless vehicles. When a vehicle is driving on a structured road, it often encounters negative obstacles such as trenches dug deep on the road due to temporary construction. If not effectively detected, there will be a risk of falling.

[0003] Traditional methods for negative obstacle detection include detection methods based on thermal infrared images, color images, and binocular vision. Among them, the detection method based on thermal infrared images detects negative obstacles in the environment according to the temperature difference between the environment and the negative obstacles. This method is easily affected by the environmental temperature; the detection method based on color images detects negative obstacles in the environment according to the color module and geometric model. This method is easily affected by the light in the environment; the detection method based on binocular vision is also easily affected by environmental light and weather changes.

[0004] To solve the problem that traditional image detection is easily affected by light and environment, lidar is usually used for negative obstacle detection in driverless vehicles. It can accurately describe the surrounding environment through point cloud data, and has the advantages of a wide detection range, high accuracy, and being unaffected by light conditions.

[0005] The invention patent application with the publication number CN112505724A and the invention name "Road Negative Obstacle Detection Method and System" superimposes single-frame environmental point cloud data according to the vehicle movement information to obtain multi-frame fused point cloud data; performs negative edge extraction on the multi-frame fused point cloud data to obtain a negative edge curve, and then judges the negative obstacle and its specific position to complete the detection of road negative obstacles. In this method, a high-density local map is obtained through multi-frame fusion, and then the position of the negative obstacle is determined through the occlusion relationship. This invention needs to use blind compensation lidars installed on both sides of the vehicle to obtain the environmental information on both sides of the road, plus the top lidar to obtain lidar point cloud data, and uses the idea of multi-frame fusion to eliminate the traditional perception blind area and achieve the design goal of logically no blind area for negative obstacle extraction. On the one hand, the above invention increases the overall hardware procurement cost because two multi-line blind compensation radars are used on both sides of the vehicle to enhance perception; on the other hand, due to certain errors in radar measurement itself, data preprocessing is required to align the positions of single-frame data before negative obstacle extraction through the fusion method, and the overall alignment of data is still a major problem in fusion. Therefore, the proposed solution of this invention has the disadvantages of cumbersome processing and poor applicability.

[0006] The invention patent application with the publication number CN106650640A and the invention name "A Negative Obstacle Detection Method Based on Local Structural Features of LiDAR Point Clouds". Two vertically installed radars are used on both sides to sense the front environment. The collected point cloud data is detected line by line for three structural features: local point cloud distance jump, local point cloud distribution density, and local point cloud height drop. Candidate point pairs that may belong to negative obstacles are extracted and screened from the single-line LiDAR point cloud based on these three structural features. All candidate point pairs obtained from the laser point cloud of each line of laser are clustered according to the consistency of point pair length and spatial position to obtain a candidate negative obstacle area, and then the negative obstacle area is obtained through area filtering and point pair number filtering. This invention completely extracts candidate negative obstacles based on the change characteristics of scan lines, and then uses means such as verifying spatial continuity information such as length and width in the post-processing stage for determination. In scenarios with a large number of moving targets (especially low-lying moving targets), there will be a problem of too high false detection rate and the stability cannot be guaranteed. Summary of the Invention

[0007] The technical problem to be solved by the present invention lies in: aiming at the problems existing in the prior art, the present invention provides a negative obstacle detection method, system, medium and device applied to structured roads, which can improve the stability and robustness of negative obstacle detection, and improve the detection accuracy and precision.

[0008] To solve the above technical problems, the technical solution proposed by the present invention is:

[0009] A negative obstacle detection method applied to structured roads, including the steps of:

[0010] Obtain the original point cloud information of the environmental scene;

[0011] Construct a ground model based on the original point cloud information, and extract ground information through the ground model;

[0012] Use the extracted ground information as prior information, and perform gradient characteristic analysis and geometric distribution statistical analysis on the rasterized adjacent point clouds beam by beam. The analysis results within the sensing range of each beam are used as candidate negative obstacle grids;

[0013] Summarize the candidate negative obstacle grids of all beams and perform filtering to output the final negative obstacle detection result.

[0014] As a further improvement of the above technical solution:

[0015] The specific process of extracting ground information through the ground model is:

[0016] Convert the rasterized map point cloud to the polar coordinate system;

[0017] Select a part of the area in front of the radar, and the heights of all the point clouds within the selected area satisfy a certain height value;

[0018] Use the average height of the point clouds in the selected area as the heuristic ground height empirical value. After obtaining the ground height empirical value, estimate the ground height of the point clouds in each angular direction in the order of angles.

[0019] The process of estimating the ground height of the point clouds in each angular direction is as follows:

[0020] First, calculate the ground empirical threshold. The rule is to increase the relaxation amount by 10 cm for every 2 m of radial distance. The ground height threshold of the target sector is obtained by adding the relaxation amount to the ground empirical threshold;

[0021] Extract the point clouds in the sector whose heights do not exceed the ground height threshold of the sector as valid ground point clouds, obtain all the valid ground point clouds in this angular direction, and finally use RANSAC to apply polynomial curve fitting in this angular direction to obtain the ground height of the sector;

[0022] Finally, obtain the ground heights of all sectors.

[0023] The process of obtaining candidate negative obstacle grids is as follows: According to the divided grids, calculate the reference ground height difference between the point clouds in the grid and the ground model for each beam in turn. If the height difference is greater than a certain value, the number of points is greater than n, and the degree of height change of these point clouds in the grid as the statistical unit is greater than a certain degree, then it is marked as a candidate negative obstacle grid.

[0024] Use the spatial neighborhood relationship for grid airspace filtering; among them, grid airspace filtering is to select the local maximum value of the grid adjacent point spacing by detecting the local grid distance jump feature as the judgment result.

[0025] Gradient characteristic analysis and geometric distribution statistical analysis include analyzing the original point cloud height difference, height change trend, height variance, point cloud smooth points, and point cloud density in the grid.

[0026] Use a horizontally installed radar to detect the front road without dead angles to obtain the original point cloud information of the environmental scene.

[0027] The present invention also discloses a negative obstacle detection system applied to a structured road, including:

[0028] A lidar point cloud acquisition module for acquiring the original point cloud information of the environmental scene;

[0029] A ground information extraction module for constructing a ground model based on the original point cloud information and extracting ground information through the ground model;

[0030] The candidate negative obstacle extraction module is used to take the extracted ground information as prior information, analyze the gradient characteristics and geometric distribution statistics of the rasterized adjacent point cloud beam by beam, and use the analysis results within the sensing range of each beam as candidate negative obstacle grids;

[0031] The negative obstacle area identification and output module is used to summarize the candidate negative obstacle grids of all beams, perform filtering, and output the final negative obstacle detection result.

[0032] The present invention also discloses a computer-readable storage medium, on which a computer program is stored, and the computer program executes the steps of the above-mentioned method when being run by a processor.

[0033] The present invention also discloses a computer device, including a memory and a processor, where a computer program is stored on the memory, and the computer program executes the steps of the above-mentioned method when being run by the processor.

[0034] Compared with the prior art, the advantages of the present invention are as follows:

[0035] (1) The negative obstacle detection method of the present invention applied to structured roads, by introducing the ground information of the ground model, compared with the methods of directly scanning line features or directly rasterizing and counting, by extracting the ground height as prior knowledge, can resist the influence of non-uniform ground materials and eliminate the statistical jumps caused by local coupling of point clouds; generally, negative obstacles in the ground have the height statistical characteristic of sinking relative to the ground plane, so stable ground height estimation can ensure the stability and robustness of negative obstacle extraction.

[0036] (2) The present invention obtains candidate negative obstacle grids through gradient characteristic analysis and geometric distribution statistical analysis, and then obtains the final negative obstacle valid points through raster airspace filtering based on spatial proximity relationships, which can reduce false alarms and improve detection accuracy.

[0037] (3) The present invention can realize the extraction of ground information and the summary of raster statistical information by traversing the original data once, can meet the real-time calculation requirements of embedded computing devices, and realize the function of sensing negative obstacles in structured roads for driverless vehicles. Description of the Drawings

[0038] Figure 1 It is a flowchart of the method of the present invention in an embodiment.

[0039] Figure 2 It is a schematic diagram of the polar coordinate ground model in the present invention.

[0040] Figure 3 It is a schematic diagram of the negative obstacle point cloud of a multi-line lidar in the present invention (the blank area in the lower left corner is the negative obstacle).

[0041] Figure 4 This is a block diagram of the system of the present invention in an embodiment. Detailed implementation manners

[0042] The present invention will be further described below in conjunction with the accompanying drawings of the specification and specific embodiments.

[0043] As Figure 1 shown, the negative obstacle detection method applied to a structured road in an embodiment of the present invention includes the steps of:

[0044] Obtain the original point cloud information of the environmental scene;

[0045] Construct a ground model based on the original point cloud information, and extract ground information through the ground model;

[0046] Use the extracted ground information as prior information, perform gradient characteristic analysis and geometric distribution statistical analysis on the rasterized adjacent point clouds beam by beam, and use the analysis results within the sensing range of each beam as candidate negative obstacle grids;

[0047] Summarize the candidate negative obstacle grids of all beams, and perform filtering to output the final negative obstacle detection result.

[0048] In a specific embodiment, the process of extracting ground information through the ground model is as follows: Convert the rasterized map point cloud to a polar coordinate system, where the angular resolution is 2 degrees and the radial distance resolution is 50 cm. Select a part of the area in front of the radar, and the height of all point clouds within the selected area satisfies: the absolute value of the height is not higher than 30 cm; use the average height of the point clouds in the selected area as the heuristic ground height empirical value. After obtaining the ground height empirical value, take the angle as the processing order, and perform ground height estimation on the point clouds in each angular direction: First, calculate the ground empirical threshold. The rule is to add a relaxation amount of 10 cm for every 2 m of radial distance. The ground empirical threshold plus the relaxation amount can obtain the ground height threshold of the target sector. Extract the point clouds in this sector whose height does not exceed the ground height threshold of the sector as valid ground point clouds, so that all valid ground point clouds in this angular direction can be obtained; finally, use RANSAC to apply polynomial curve fitting in this angular direction, and the fitting coefficient is 3 times (or more times); perform this operation method on each angular direction to obtain the ground height of all sectors.

[0049] Compared with the method of directly scanning line features or directly rasterizing and dividing statistics, introducing a ground model to extract the ground height as prior knowledge can thus resist the influence of non-uniform ground materials and eliminate statistical jumps caused by local coupling of point clouds. Negative obstacles on the ground generally have the height statistical characteristic of sinking downward relative to the ground plane. Therefore, stable ground height estimation can ensure the stability and robustness of negative obstacle extraction.

[0050] In a specific embodiment, after determining the candidate negative obstacle grid, grid airspace filtering is further performed using the spatial proximity relationship; wherein the grid airspace filtering selects the local maximum value of the grid adjacent point spacing as the determination result by detecting the local grid distance jump feature. Specifically, since there is a distance jump in the point cloud for negative obstacles, and the height after the jump is lower than the ground model reference height. Therefore, using the above grid airspace filtering can effectively remove noise points on the one hand, and accurately find the negative obstacle area according to the characteristics of local distance jump on the other hand. The above filtering based on the spatial neighborhood relationship is used to eliminate isolated false alarm points, is applicable to targets with a certain regular geometric shape, and maintains continuity in the spatial relationship.

[0051] In a specific embodiment, the gradient characteristic analysis and geometric distribution statistical analysis include analyzing the height difference, height change trend, height variance, point cloud smoothness points, and point cloud density of the original point cloud within the grid. Specifically, the calculation method of the height difference is: calculating the difference between the height of the point cloud and the height output by the ground reference model; the calculation method of the height change rate is: calculating the slope of the point in the height direction and converting it into an angle value of 0 - 180 degrees. The threshold condition for the negative obstacle boundary and the ground is: the angle between adjacent grids is greater than or equal to 30 degrees; the calculation method of the point cloud smoothness is: statistically calculating the variance of the point cloud belonging to the same grid; the calculation method of the point cloud density is: statistically calculating the number of point clouds in the same grid, and converting the point cloud values into four levels of very sparse, sparse, medium density, and high density for unified description.

[0052] In a specific embodiment, the location information includes the ground equation and the current ground height information (due to the accumulation of various errors, the ground height obtained from the original laser point cloud is not absolutely 0, and the ground model can accurately find the ground height information in this area).

[0053] In a specific embodiment, a horizontally installed radar is used to detect the front road without dead angles to obtain the original point cloud information of the environmental scene. By innovatively transforming the radar installation method to horizontal installation, blind - area - free perception of the front road of the vehicle is achieved.

[0054] The negative obstacle detection method of the present invention applied to structured roads, by introducing the ground information of the ground model, compared with the method of directly scanning line features or directly dividing grids for statistics, by extracting the ground height as prior knowledge, can resist the influence of non - homogeneous ground materials and eliminate the statistical jumps caused by local coupling of the point cloud; generally, negative obstacles on the ground have the height statistical characteristic of sinking downward relative to the ground plane. Therefore, stable ground height estimation can ensure the stability and robustness of negative obstacle extraction.

[0055] The present invention obtains candidate negative obstacle grids through gradient characteristic analysis and geometric distribution statistical analysis, and then performs grid airspace filtering through spatial proximity relationships to obtain the final effective negative obstacle points, which can reduce false alarms and improve detection accuracy.

[0056] The present invention can realize the extraction of ground information and the summary of grid statistical information by traversing the original data once, can meet the real-time calculation requirements of embedded computing devices, and realize the function of a driverless vehicle to sense negative obstacles on a structured road. As Figure 4 shown, an embodiment of the present invention also discloses a negative obstacle detection system applied to a structured road, including:

[0057] A lidar point cloud acquisition module for acquiring the original point cloud information of the environmental scene;

[0058] A ground information extraction module for constructing a ground model based on the original point cloud information and extracting ground information through the ground model;

[0059] A candidate negative obstacle extraction module for using the extracted ground information as prior information, performing gradient characteristic analysis and geometric distribution statistical analysis on the rasterized adjacent point clouds beam by beam, and taking the analysis results within the sensing range of each beam as candidate negative obstacle grids;

[0060] A negative obstacle area identification and output module for summarizing the candidate negative obstacle grids of all beams and performing filtering to output the final negative obstacle detection result.

[0061] The system of the present invention corresponds to the above method and has the same advantages as described in the above method.

[0062] The following further illustrates the present invention in conjunction with a complete specific embodiment:

[0063] Use a radar to scan the surrounding environment to obtain a three-dimensional point cloud description of the surrounding scene, as Figure 3 shown; among them, the point cloud data is obtained by a horizontally installed lidar, and this installation method enables the radar to detect the front road without dead angles; the lidar operates in a multi-line laser rotation scanning mode, and one laser beam corresponds to a continuously distributed point cloud; of course, single-line laser point cloud can also be used to replace the multi-line laser point cloud;

[0064] Generate a ground description based on the ground model algorithm and extract ground information. Among them, the ground model is dynamically generated based on the collected surrounding environment point cloud; the ground information includes the ground equation and the current ground height information;

[0065] Specifically, the process of the ground model extracting ground information is as follows: The rasterized map point cloud is transformed into the polar coordinate system, where the angular resolution is 2 degrees and the radial distance resolution is 50 cm. A part of the area is selected from the front area of the radar, and the heights of all the point clouds within the selected area satisfy: the absolute value of the height is not higher than 30 cm; the average height of the point clouds in the selected area is used as the heuristic ground height empirical value. After obtaining the ground height empirical value, taking the angle as the processing order, the ground height of the point clouds in each angular direction is estimated: First, calculate the ground empirical threshold. The rule is that the relaxation amount increases by 10 cm for every 2 m in the radial distance. The ground height threshold of the target sector can be obtained by adding the relaxation amount to the ground empirical threshold. The point clouds within the sector whose heights do not exceed the ground height threshold of the sector are extracted as valid ground point clouds, and in this way, all the valid ground point clouds in this angular direction can be obtained; Finally, in this angular direction, RANSAC is used to apply polynomial curve fitting, and the fitting coefficient is 3 times (or more). By using this operation method for each angular direction, the ground heights of all sectors can be obtained. The calculation result of the ground model is the ground height and the polynomial fitting coefficient in this angular direction.

[0066] As Figure 3 shown, taking the vehicle as the origin, the Figure 3 multi-beam point cloud data shown is divided into different grids according to the Figure 2 polar coordinates shown, and the ground information is obtained by calculating the ground model within each grid; Then, the point clouds of each beam are rasterized and clustered, and the extracted ground information is used as prior information for analysis. Specifically, the gradient characteristic analysis and geometric distribution statistical analysis are carried out on the rasterized adjacent point clouds beam by beam. The analysis results within the sensing range of each beam are used as the first negative obstacle determination result and output as candidate negative obstacle grids; The gradient characteristic analysis refers to the height difference, change rate and curvature with respect to the reference ground, specifically the first and second derivatives in the Z direction; The geometric distribution statistical analysis refers to the point cloud density and height variance in the statistical direction; Specifically, the process of obtaining the candidate negative obstacle grids is as follows: According to the divided grids, calculate the height difference between the point clouds in the grid and the reference ground of the ground model beam by beam. If the number of points lower than the reference height of the ground model by 30 cm is greater than 10, and the degree of height change of these points with the grid as the statistical unit is greater than 30 degrees, it is marked as a candidate negative obstacle grid.

[0067] Further, the candidate negative obstacle grids of all wire bundles are summarized, and grid airspace filtering is performed using spatial proximity relationships, and the output result is used as the second negative obstacle determination result; wherein the grid airspace filtering selects the local maximum value of the grid adjacent point spacing as the determination result by detecting the local grid distance jump feature. Specifically, the point clouds within adjacent grids are summarized, the local grid distance jump feature is detected. If the distance jump is greater than 10 cm and there are candidate negative obstacle points in the grid and other candidate negative obstacle points can be found within 5 cm nearby, it is marked as a valid negative obstacle point. The negative obstacle area output module summarizes the second negative obstacle determination results of all wire bundles as the final negative obstacle output.

[0068] The present invention also discloses a computer-readable storage medium, on which a computer program is stored, and the computer program executes the steps of the above-mentioned method when being run by a processor. The present invention further discloses a computer device, including a memory and a processor, a computer program is stored on the memory, and the computer program executes the steps of the above-mentioned method when being run by the processor.

[0069] The implementation of all or part of the processes in the above embodiment methods of the present invention can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal, and software distribution medium, etc. The memory can be used to store computer programs and / or modules. The processor realizes various functions by running or executing the computer programs and / or modules stored in the memory, and calling the data stored in the memory. The memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one magnetic disk storage device, flash device, or other volatile solid-state storage devices, etc.

[0070] The above are only the preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the concept of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art, several improvements and refinements made without departing from the principle of the present invention should be regarded as within the protection scope of the present invention.

Claims

1. A negative obstacle detection method applied to structured roads, characterized in that, Including the steps: Obtain the original point cloud information of the environmental scene; Construct a ground model based on the original point cloud information, and extract ground information through the ground model; Take the extracted ground information as prior information, perform gradient characteristic analysis and geometric distribution statistical analysis on the rasterized adjacent point clouds beam by beam, and take the analysis results within the sensing range of each beam as candidate negative obstacle rasters; Summarize the candidate negative obstacle rasters of all beams, and perform filtering to output the final negative obstacle detection result; The specific process of extracting ground information through the ground model is: Convert the rasterized map point cloud to the polar coordinate system; Select a part of the area from the front area of the radar, and the heights of all point clouds within the selected area satisfy a certain height value; Use the average height of the point clouds in the selected area as the heuristic ground height empirical value. After obtaining the ground height empirical value, estimate the ground height of the point clouds in each angular direction in the order of angles; The process of estimating the ground height of the point clouds in each angular direction is: First, calculate the ground empirical threshold. The rule is to add a relaxation amount of 10 cm for every 2 m of radial distance. The ground height threshold of the target sector is obtained by adding the relaxation amount to the ground empirical threshold; Extract the point clouds within the sector whose heights do not exceed the ground height threshold of the sector as valid ground point clouds, obtain all valid ground point clouds in this angular direction, and finally use RANSAC to apply polynomial curve fitting in this angular direction to obtain the ground height of the sector; Finally, obtain the ground heights of all sectors.

2. The negative obstacle detection method applied to a structured road according to claim 1, characterized in that The process of obtaining candidate negative obstacle rasters is: According to the divided rasters, calculate the height difference between the point clouds in the raster and the reference ground height of the ground model beam by beam. If the height difference is greater than a certain value and the number of points is greater than n, and the degree of height change of these point clouds with the raster as the statistical unit is greater than a certain degree, then mark it as a candidate negative obstacle raster.

3. The negative obstacle detection method applied to a structured road according to any one of claims 1 to 2, characterized in that, Perform raster airspace filtering using spatial neighborhood relationships; among them, raster airspace filtering selects the local maximum value of the raster adjacent point spacing by detecting the local raster distance jump feature as the determination result.

4. The negative obstacle detection method applied to a structured road according to any one of claims 1 to 2, characterized in that, The gradient characteristic analysis and geometric distribution statistical analysis include analyzing the original point cloud height difference, height change trend, height variance, point cloud smooth points, and point cloud density within the raster.

5. The negative obstacle detection method applied to a structured road according to any one of claims 1 to 2, characterized in that, Use a horizontally installed radar to detect the front road without dead angles to obtain the original point cloud information of the environmental scene.

6. A negative obstacle detection system applied to a structured road, which is used to execute the steps of the negative obstacle detection method applied to a structured road according to any one of claims 1-5, characterized in that, Including: A lidar point cloud acquisition module for obtaining the original point cloud information of the environmental scene; A ground information extraction module for constructing a ground model based on the original point cloud information and extracting ground information through the ground model; A candidate negative obstacle extraction module for taking the extracted ground information as prior information, performing gradient characteristic analysis and geometric distribution statistical analysis on the rasterized adjacent point clouds beam by beam, and taking the analysis results within the sensing range of each beam as candidate negative obstacle rasters; A negative obstacle area identification and output module for summarizing the candidate negative obstacle rasters of all beams and performing filtering to output the final negative obstacle detection result.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program, when run by a processor, executes the steps of the method according to any one of claims 1 to 5.

8. A computer device, comprising a memory and a processor, wherein a computer program is stored on the memory, characterized in that, The computer program, when run by a processor, performs the steps of the method according to any one of claims 1 to 5.

Images