An unmanned forklift road perception method and system based on angle density and structure constraint

Abstract

The application discloses an unmanned pushing vehicle road perception method and system based on angle density and structure constraint, and the method comprises the following steps: S1, point cloud preprocessing: obtaining a laser radar original point cloud, filtering out invalid points, ground height abnormal points, too high object height abnormal points and point clouds near the vehicle itself, and obtaining an effective two-dimensional point cloud set for perception; S2, road direction rough estimation based on angle density; S3, precise detection of railing boundary based on structure constraint; S4, result output and tracking. The application has stronger anti-interference ability to noise and temporary obstacles in the point cloud by combining angle density macroscopic statistics and microscopic transverse gradient search, and is more widely used; the scene features are used as strong prior geometric constraints to verify the physical rationality of the detection result, so that false detection is avoided, and the accuracy and reliability of boundary positioning are improved; and the 180-degree ambiguity problem of long straight road direction judgment is solved.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving environmental perception technology, specifically a road perception method and system for an unmanned material pusher based on angle density and structural constraints. Background Technology

[0002] In indoor or enclosed environments where GNSS is unavailable, autonomous vehicles typically rely on lidar or visual sensors for environmental perception. Existing technological solutions mainly include:

[0003] 1. Road detection based on point cloud clustering: Ground segmentation and clustering of point clouds are performed, and larger, linear clusters are identified as road boundaries or guide lines. However, the effect is unstable when there are many obstacles or irregular boundaries.

[0004] 2. Positioning and matching based on pre-built high-precision maps: The indoor environment is surveyed and mapped with high precision in advance, and a point cloud map is built. Real-time positioning is achieved through point cloud registration. However, it relies on high-precision maps, lacks flexibility, and has high map maintenance costs.

[0005] The existing technologies mentioned above have the following problems when dealing with typical indoor structured environments with certain characteristics: clustering-based or general fitting methods are sensitive to temporary obstacles, dynamic objects, and sensor noise in the environment, which can easily lead to abrupt changes or loss of road or boundary detection results; existing general perception algorithms usually do not explicitly utilize the strong geometric prior knowledge of road and scene features, making it difficult to accurately locate the true road boundaries in complex point clouds. Therefore, to address the above problems, a road perception method and system for unmanned material pushers based on angle density and structural constraints is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a road perception method and system for unmanned material pushers based on angle density and structural constraints in order to solve the above-mentioned problems.

[0007] This invention achieves the above objective through the following technical solution: a road perception method for an unmanned material pusher based on angle density and structural constraints, comprising the following steps: S1. Point cloud preprocessing: Obtain the original point cloud of the lidar, filter out invalid points, points with abnormal heights, and point clouds near the vehicle itself, and obtain an effective two-dimensional point cloud set for perception. S2. Coarse road direction estimation based on angle density: The coarse road direction estimation is obtained by angle normalization, density statistics and smoothing, low-density interval detection and merging, and direction determination and tracking, to obtain the most reliable road direction angle θ_road in the current frame by priority sorting. S3. Fine-grained detection of railing boundaries based on structural constraints: The detection process for fine-grained railing boundaries is as follows: S301, Coordinate Transformation: Based on θ_road determined in step S2, rotate all point clouds to the road coordinate system so that the x-axis is parallel to the road direction and the y-axis is perpendicular to the road. S302, Lateral density gradient search: In the road coordinate system, divide the interval along the y-axis within a certain range and count the lateral point cloud density; S303, Geometric Constraint Verification: Using the prior knowledge of the known railing spacing, calculate y_right - y_left and verify whether it is within the tolerance range. If the verification passes, the railing position is confirmed to be valid. S304. Boundary line fitting: Based on the confirmed horizontal position of the railing, extract the point cloud clusters belonging to the left and right railings from the original point cloud, and fit them as straight lines in the original two-dimensional plane to obtain the mathematical equations of the left and right boundary lines. S4. Output and Tracking: Output the road direction angle θ_road and the equations of the left and right boundary lines of the current frame, update the historical tracking status, and provide prior information for the next frame processing.

[0008] A further technical solution involves a rough estimation of the road direction, with the following specific steps: S201, Angle Normalization: Convert the preprocessed two-dimensional point cloud into polar coordinates, and treat angles θ and θ+π as the same direction, normalizing all angles to the interval [0, π) to resolve directional ambiguity; S202, Density Statistics and Smoothing: The point cloud angle density is statistically analyzed within the interval [0, π) at a set resolution, and Gaussian filtering is used for smoothing to obtain a smoothed angle density distribution map. S203, Low-density interval detection and merging: Set a threshold to identify low-density continuous intervals in the density distribution map, which correspond to the direction of open roads; innovatively, if there is an interval that ends near 180° and another interval that starts near 0°, then merge the two into a continuous low-density interval that spans the 0° / 180° boundary. Further technical solutions, S204, Direction determination and tracking: Evaluate the detected low-density areas and combine them with historically detected road direction information.

[0009] A further technical solution, the lateral density gradient search in step S302: in the road coordinate system, divide the interval along the y-axis within the range of [-4.3m, 4.3m], and count the lateral point cloud density; Calculate the density gradient and search from near y=0 in the positive (right) and negative (left) y directions respectively to find the first consecutive position that satisfies the condition that the density is significantly higher than the background threshold and the gradient is positive (right) or negative (left), and use it as the initial estimated position (y_left, y_right) of the left and right rails.

[0010] A further technical solution, the geometric constraint verification in step S303: using the prior knowledge of the known railing spacing W, not limited to W=4.85m, calculate y_right - y_left, and verify whether it is within the tolerance range: W±0.5m. If the verification passes, the railing position is confirmed to be valid.

[0011] In a further technical solution, the mathematical equations of the left and right boundary lines are not limited to the one known geometric relationship y=kx+b.

[0012] A road perception system for an unmanned material pusher based on angle density and structural constraints, characterized in that it includes sensors, a point cloud preprocessing module, a road direction detection module, and a guardrail detection module; The point cloud preprocessing module is used to filter out invalid points, perform height filtering, remove vehicle point clouds, and filter out outliers from the acquired raw point cloud of the lidar. The road direction detection module is used to detect coordinate transformations, angle normalization, angle density statistics and smoothing, low-density interval detection and merging, road direction calculation, historical tracking and confidence management, and multi-candidate decision-making in historical direction tracking. The railing detection module is used to detect and verify the prior conditions of the structure: coordinate transformation, lateral density statistics, Gaussian smoothing, density gradient calculation, search for the first high-density transition point, prior condition verification, railing point cloud extraction, and line fitting.

[0013] In a further technical solution, the sensor is not limited to a 16-line lidar, but is installed on the top of the vehicle.

[0014] Compared with the prior art, the beneficial effects of the present invention include: I. High robustness: By combining macroscopic statistics of angle density and microscopic lateral gradient search, it has stronger anti-interference ability against noise and temporary obstacles in point clouds, and improves the stability of perception in complex indoor environments.

[0015] II. High precision and strong constraints: By using scene features as strong prior geometric constraints, the physical rationality of the detection results is verified, avoiding false detections and improving the accuracy and reliability of boundary positioning.

[0016] Third, effectively solve the ambiguity of direction: By periodically normalizing the angle (θ is equivalent to θ+π) and merging the intervals that cross the boundary, the 180-degree ambiguity problem of direction judgment for long straight roads is solved.

[0017] IV. High computational efficiency, suitable for real-time applications: The algorithm process is clear, avoiding complex point cloud global registration or large-scale deep learning model inference, with low computational burden, and can run in real time on the vehicle computing unit. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the 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 invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0020] Figure 1 This is a diagram of the overall architecture of the present invention; Figure 2 This is an angular density distribution diagram of the present invention; Figure 3 This is a schematic diagram of the low-density region of the present invention; Figure 4 This is a schematic diagram of the lateral density gradient search in the road coordinate system of this invention. Detailed Implementation

[0021] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0022] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0023] In the description of this invention, it should be understood that the terms "upper", "lower", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0024] Please see Figure 1-4 As shown, a road perception method for an unmanned material pusher based on angle density and structural constraints includes the following steps: S1. Point cloud preprocessing: Obtain the original point cloud of the lidar, filter out invalid points, ground height anomalies, height anomalies of excessively tall objects, and point clouds near the vehicle itself, and obtain an effective two-dimensional point cloud set for perception. S2. Coarse estimation of road direction based on angle density, the specific process is as follows: S201, Angle Normalization: Convert the preprocessed two-dimensional point cloud into polar coordinates, and treat angles θ and θ+π as the same direction, normalizing all angles to the interval [0, π) to resolve directional ambiguity; S202, Density Statistics and Smoothing: Point cloud angle density is statistically analyzed within the interval [0, π) at a set resolution of 1°, and then smoothed using Gaussian filtering to obtain a smoothed angle density distribution map, as shown below. Figure 2 ; S203, Low-density region detection and merging: Set a threshold, such as Figure 3 It identifies low-density continuous intervals in the density distribution map, which correspond to open road directions; innovatively, if there is an interval that ends near 180° and another interval that begins near 0°, the two are merged into a continuous low-density interval that spans the 0° / 180° boundary. This paper proposes a method for handling angular periodicity and interval merging. The radar point cloud angles θ and θ+π are normalized to the same interval [0, π) for density statistics. A mechanism is designed to identify and merge low-density intervals that cross the 0° / 180° boundary, which effectively solves the directional deviation problem of long straight roads and ensures the accuracy and continuity of directional tracking.

[0025] S204, Direction Determination and Tracking: Evaluate the detected low-density areas and combine them with historically detected road direction information.

[0026] The lateral density gradient search described in step S302 involves dividing the area into intervals along the y-axis within the range of [-4.3m, 4.3m] in the road coordinate system, and statistically analyzing the lateral point cloud density, as shown below. Figure 4 ; Calculate the density gradient and search from near y=0 in the positive (right) and negative (left) y directions respectively to find the first consecutive position that satisfies the condition that the density is significantly higher than the background threshold and the gradient is positive (right) or negative (left). Use this position as the initial estimated position (y_left, y_right) of the left and right rails. Then, obtain the most reliable road direction angle θ_road for the current frame by prioritizing the selection. This paper proposes a search strategy based on the "first significant transition point" of the lateral density gradient. In the road coordinate system, instead of relying on finding the global highest density peak, it searches for the first transition point that satisfies the density and gradient thresholds from the center of the road to both sides. This can effectively eliminate the interference of temporary obstacles on the road and more reliably locate the true fixed guardrail boundary.

[0027] S3. Fine-grained detection of railing boundaries based on structural constraints: The detection process for fine-grained railing boundaries is as follows: S301, Coordinate Transformation: Based on θ_road determined in step S2, rotate all point clouds to the road coordinate system so that the x-axis is parallel to the road direction and the y-axis is perpendicular to the road. S302, Lateral density gradient search: In the road coordinate system, divide the interval along the y-axis within a certain range and count the lateral point cloud density; S303, Geometric Constraint Verification: Using the prior knowledge of the known railing spacing W, not limited to W=4.85m, calculate y_right - y_left and verify whether it is within the tolerance range: W±0.5m. If the verification passes, the railing position is confirmed to be valid. S304. Boundary line fitting: Based on the confirmed horizontal position of the railing, extract the point cloud clusters belonging to the left and right railings from the original point cloud, and fit them as straight lines in the original two-dimensional plane to obtain the mathematical equations of the left and right boundary lines y=kx+b. This paper proposes an indoor structured road perception framework that integrates macroscopic statistics of angle density with prior geometric constraints. The framework robustly determines the macroscopic direction of the road through angle density analysis, and then uses the known parallel boundary spacing as a strong constraint to accurately locate the road in the direction perpendicular to the road. This enables stable and accurate detection of road boundaries under complex interference.

[0028] S4. Output and Tracking: Output the road direction angle θ_road and the equations of the left and right boundary lines of the current frame, update the historical tracking status, and provide prior information for the next frame processing.

[0029] like Figure 1 As shown, an unmanned material pusher road perception system based on angle density and structural constraints is characterized by including sensors, a point cloud preprocessing module, a road direction detection module, and a guardrail detection module. The point cloud preprocessing module is used to filter out invalid points, perform height filtering, remove vehicle point clouds, and filter out outliers from the acquired raw point cloud of the lidar. The road direction detection module is used to detect coordinate transformations, angle normalization, angle density statistics and smoothing, low-density interval detection and merging, road direction calculation, historical tracking and confidence management, and multi-candidate decision-making in historical direction tracking. The railing detection module is used to detect and verify the prior conditions of the structure: coordinate transformation, lateral density statistics, Gaussian smoothing, density gradient calculation, search for the first high-density transition point, prior condition verification, railing point cloud extraction, and line fitting. The sensor is not limited to a 16-line lidar and is installed on the top of the vehicle.

[0030] Example 1 A road perception method and system for an unmanned material pusher based on angle density and structural constraints is described below: The first step is system initialization and parameter settings.

[0031] Sensor: A 16-line lidar is used and installed on the roof of the vehicle; Parameters: Set the prior value of road railing spacing W=4.85m, angle resolution 1°, set the low density interval length range of 11°-55° according to the road width, the lateral search range y∈[-4.3, 4.3]m, and the spacing tolerance 0.5m.

[0032] The second step is the online sensing process.

[0033] 1. Preprocessing: Receive one frame of point cloud. Filter out points with heights less than -0.6m (ground) and greater than 0.2m (top obstacles), and filter out points within 1.0m of the radar center (vehicle body), obtaining approximately 8000 valid points; 2. Road Direction Detection: Valid points are projected onto the XY plane, polar coordinates are calculated, and angles are normalized. After statistically analyzing angle density, a low-density interval [175°, 180°] and another [0°, 12°] are identified. The system merges these into the interval [175°, 192°], calculates the center direction as 3.5° (i.e., 183.5°-180°), and combines this with the historical direction from the previous frame (2.1°), confirming the current road direction as θ_road = 3.5°. 3. Railing Detection: Rotate all point clouds by -3.5° and statistically analyze the density along the y-axis. Searching to the right from y=0, a density jump point (gradient > 5) is detected at y=2.41m; searching to the left, a jump point (gradient < -5) is detected at y=-2.39m. The calculated spacing d=4.80m has an error of 0.05m compared to W=4.85m, which is less than the tolerance of 0.5m, thus the verification is successful. Extract the point clouds for y∈[2.0, 2.8] and y∈[-2.8, -2.0], and fit them to obtain the left railing line: y = 0.061x - 2.42, and the right railing line: y = 0.058x + 2.40, respectively. 4. Output: Publish the road direction at 3.5° and the straight-line equations of the left and right guardrails for use by downstream control modules. It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalent elements of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0034] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A road perception method for an unmanned material pusher based on angle density and structural constraints, characterized in that: Includes the following steps: S1. Point cloud preprocessing: Obtain the original point cloud of the lidar, filter out invalid points, points with abnormal heights, and point clouds near the vehicle itself, and obtain an effective two-dimensional point cloud set for perception. S2. Coarse road direction estimation based on angle density: The coarse road direction estimation is obtained by angle normalization, density statistics and smoothing, low-density interval detection and merging, and direction determination and tracking, to obtain the most reliable road direction angle θ_road in the current frame by priority sorting. S3. Fine-grained detection of railing boundaries based on structural constraints: The detection process for fine-grained railing boundaries is as follows: S301, Coordinate Transformation: Based on θ_road determined in step S2, rotate all point clouds to the road coordinate system so that the x-axis is parallel to the road direction and the y-axis is perpendicular to the road. S302, Lateral density gradient search: In the road coordinate system, divide the interval along the y-axis within a certain range and count the lateral point cloud density; S303, Geometric Constraint Verification: Using the prior knowledge of the known railing spacing, calculate y_right - y_left and verify whether it is within the tolerance range. If the verification passes, the railing position is confirmed to be valid. S304. Boundary line fitting: Based on the confirmed horizontal position of the railing, extract the point cloud clusters belonging to the left and right railings from the original point cloud, and fit them as straight lines in the original two-dimensional plane to obtain the mathematical equations of the left and right boundary lines. S4. Output and Tracking: Output the road direction angle θ_road and the equations of the left and right boundary lines of the current frame, update the historical tracking status, and provide prior information for the next frame processing.

2. The road perception method for an unmanned material pusher based on angle density and structural constraints according to claim 1, characterized in that: The coarse estimation of the road direction involves the following steps: S201, Angle Normalization: Convert the preprocessed two-dimensional point cloud into polar coordinates, and treat angles θ and θ+π as the same direction, normalizing all angles to the interval [0, π) to resolve directional ambiguity; S202, Density Statistics and Smoothing: The point cloud angle density is statistically analyzed within the interval [0, π) at a set resolution, and Gaussian filtering is used for smoothing to obtain a smoothed angle density distribution map. S203, Low-density interval detection and merging: Set a threshold to identify low-density continuous intervals in the density distribution map, which correspond to the direction of open roads; innovatively, if there is an interval that ends near 180° and another interval that starts near 0°, then merge the two into a continuous low-density interval that spans the 0° / 180° boundary. S204, Direction Determination and Tracking: Evaluate the detected low-density areas and combine them with historically detected road direction information.

3. The road perception method for an unmanned material pusher based on angle density and structural constraints according to claim 1, characterized in that: The lateral density gradient search described in step S302: In the road coordinate system, divide the interval along the y-axis within the range of [-4.3m, 4.3m] and count the lateral point cloud density; Calculate the density gradient and search from near y=0 in the positive (right) and negative (left) y directions respectively to find the first consecutive position that satisfies the condition that the density is significantly higher than the background threshold and the gradient is positive (right) or negative (left), and use it as the initial estimated position (y_left, y_right) of the left and right rails.

4. The road perception method for an unmanned material pusher based on angle density and structural constraints according to claim 1, characterized in that: The geometric constraint verification in step S303: Using the prior knowledge of the known railing spacing W, which is not limited to W=4.85m, calculate y_right - y_left and verify whether it is within the tolerance range: W±0.5m. If the verification is successful, the railing position is confirmed to be valid.

5. The road perception method and system for an unmanned material pusher based on angle density and structural constraints according to claim 1, characterized in that: The mathematical equations for the left and right boundary lines are not limited to the one known geometric relationship y=kx+b.

6. A road perception system for an unmanned material pusher based on angle density and structural constraints, characterized in that: This includes sensors, a point cloud preprocessing module, a road direction detection module, and a guardrail detection module; The point cloud preprocessing module is used to filter out invalid points, perform height filtering, remove vehicle point clouds, and filter out outliers from the acquired raw point cloud of the lidar. The road direction detection module is used to detect coordinate transformations, angle normalization, angle density statistics and smoothing, low-density interval detection and merging, road direction calculation, historical tracking and confidence management, and multi-candidate decision-making in historical direction tracking. The railing detection module is used to detect and verify the prior conditions of the structure: coordinate transformation, lateral density statistics, Gaussian smoothing, density gradient calculation, search for the first high-density transition point, prior condition verification, railing point cloud extraction, and line fitting.

7. A road perception system for an unmanned material pusher based on angle density and structural constraints according to claim 6, characterized in that: The sensor is not limited to a 16-line lidar and is installed on the top of the vehicle.

Images