Three-dimensional dynamic sight distance evaluation system and method based on laser radar point cloud data
Through the three-dimensional dynamic visual line evaluation system of lidar point cloud data, the problem that the existing technology cannot accurately reflect the vehicle visual line requirements in the real driving environment is solved, and the visual line assessment in the three-dimensional environment is realized, and the road driving safety is improved sex.
Patent Information
- Application Number
- CN202111677039.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-31
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2041-12-31
AI Technical Summary
The prior art cannot accurately reflect the vehicle's visual range requirements in the real driving environment, resulting in deviations in the visual range assessment results and it is difficult to determine the effectiveness.
Using a three-dimensional dynamic visual line-of-view evaluation system based on lidar point cloud data, preprocessing and separation is performed by obtaining vehicle trajectory and point cloud data, and the three-dimensional geometric features of the road are extracted using the Delaunay triangulation algorithm and the RANSAC segmentation algorithm, combining the operating speed The prediction model calculates the driver's dynamic available sight line, and compares the driving sight line with the three-dimensional dynamic available sight line, and determines that the minimum value of the sight line is the real available sight line.
It realizes the accurate estimation of the available visual distance of the road in a three-dimensional environment, improves the safety of road driving, and provides a theoretical basis and analytical and calculation methods.
Smart Images

Figure CN114445576B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of highway safety evaluation, and in particular to a three-dimensional dynamic sight distance evaluation system and method based on laser radar point cloud data. Background Art
[0002] Due to the complex driving environment on the road, the actual operating speed of the vehicle changes continuously with the changes in road conditions, driving skills and vehicle performance. Therefore, the static two-dimensional design sight distance of the separated horizontal and vertical roads cannot accurately reflect the demand for driving sight distance in the real driving environment. At the same time, there is a certain deviation in the sight distance estimation for the road, which makes it difficult to determine the effectiveness of the sight distance assessment results. Summary of the invention
[0003] In order to overcome the shortcomings and deficiencies of the prior art, the present invention provides a three-dimensional dynamic sight distance assessment system and method based on lidar point cloud data, specifically, accurately estimating the available sight distance in the three-dimensional space of the road based on a running speed prediction model.
[0004] The present invention adopts the following technical solution:
[0005] A three-dimensional dynamic sight distance assessment method based on laser radar point cloud data comprises the following steps:
[0006] Obtain vehicle trajectory and point cloud data;
[0007] Preprocess and separate the point cloud data to form ground point cloud data and point cloud data above the ground;
[0008] The Delaunay triangulation algorithm is used to form a TIN triangulation network for the ground point cloud data, and the linear data of the road is extracted using the TIN triangulation network; the road is segmented according to the combination of linear shapes to obtain a three-dimensional geometric feature set of the road; the running speed prediction model based on the three-dimensional linear geometric features is used to obtain the running speed of the vehicle, and the dynamic viewing angle range of the driver at each moment is determined according to the running speed;
[0009] Classify the point cloud data above the ground to form homogeneous objects and non-homogeneous objects, and further obtain a three-dimensional environment containing only road information;
[0010] The most unfavorable position of visibility is when the vehicle is biased toward the edge line of one side of the one-way road, and the most favorable position of visibility is when the vehicle is biased toward the other side. The smaller of the two is the three-dimensional dynamic available visibility of the driver in a real three-dimensional environment.
[0011] Furthermore, it also includes a sight distance evaluation step, calculating the driving sight distance of the vehicle at the position and comparing it with the three-dimensional dynamic available sight distance at the position. If the three-dimensional dynamic available sight distance is greater than the driving sight distance, the requirement is met, otherwise it is not met.
[0012] Furthermore, the most unfavorable position of the sight distance is when the vehicle is biased toward the edge line of one side of the one-way road, and the most favorable position of the sight distance is when the vehicle is biased toward the other side. The smaller sight distance between the two is the three-dimensional dynamic available sight distance of the driver in a real three-dimensional environment, specifically:
[0013] When the driver is in the most unfavorable sight distance vehicle position, the farthest position of the lane edge line that the driver can see is used as the three-dimensional dynamic sight distance by using the cone ellipse surface, and the route arc length distance of the driver's current position corresponding to the lane edge line;
[0014] When the vehicle driver is in the most advantageous sight distance vehicle position, the three-dimensional dynamic sight distance is obtained in the same way as the above method;
[0015] The three-dimensional dynamic viewing distances obtained at the two positions are compared, and the smaller value is the three-dimensional dynamic available viewing distance of the driver in a real three-dimensional environment.
[0016] Furthermore, the preprocessed point cloud data is subjected to a RANSAC segmentation algorithm to form ground data and other point cloud data above the ground.
[0017] Furthermore, the dynamic viewing angle is the maximum horizontal and vertical viewing angle that the driver can see in the dynamic field of view state, and the elliptical cone of each driver's position is determined by this viewing angle, and its projection to the vertical two-dimensional space is the cone elliptical surface.
[0018] A system for a three-dimensional dynamic sight distance assessment method comprises a laser radar and an inertial navigation system, wherein the inertial navigation system comprises an inertial navigation unit and a GPS unit.
[0019] Furthermore, the laser radar is arranged on the top of the vehicle and maintained in a horizontal position.
[0020] Furthermore, the inertial navigation unit is installed on the center line of the rear wheel of the vehicle with the Y-axis direction facing the forward direction of the vehicle.
[0021] Furthermore, the timestamps of the laser radar and the inertial navigation system are synchronized, and the origin of the point cloud collection coordinate system starts at the laser radar position. The inertial navigation system provides the position of the vehicle in the earth coordinate system.
[0022] Furthermore, intra-frame correction and feature extraction are performed using the LOAM algorithm.
[0023] Beneficial effects of the present invention:
[0024] A real three-dimensional environment can be established through the point cloud data of LiDAR technology, and a three-dimensional dynamic visibility evaluation method for the road is formed through the driver's vision cone space at different running speeds, which provides a theoretical basis and analytical calculation method for dynamic visibility inspection based on three-dimensional road scenes and improving road driving safety. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 A schematic diagram of the three-dimensional dynamic available sight distance of a highway according to the present invention;
[0026] Figure 2 It is the position where the vehicle is in the most unfavorable sight distance state on the road section of the present invention.
[0027] Figure 3 It is a flow chart of the present invention.
[0028] The figure shows:
[0029] 1. Driver's viewpoint; 2. Lane centerline; 3. Lane edge line; 4. Maximum horizontal dynamic angle of view; 5. Maximum vertical dynamic angle of view; 6. Viewpoint centerline; 7. Position of target object; 8. Roadside obstacle area; 9. Vehicle position with the most unfavorable sight distance; 10. Lane dividing line; 11. Lane edge line; 12. Road markings; 13. Obstacle area. DETAILED DESCRIPTION
[0030] The present invention will be further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.
[0031] Example
[0032] like Figure 1-Figure 3 As shown, a three-dimensional dynamic sight distance assessment system and method based on laser radar point cloud data, the system part includes laser radar and inertial navigation system, and the inertial navigation system includes inertial navigation unit and GPS. The timestamps of laser radar and inertial navigation system are synchronized, and intra-frame correction and feature extraction are performed through LOAM algorithm. At the same time, the origin of the point cloud coordinate system starts at the position of laser radar, so that the inertial navigation system and laser radar can provide the trajectory of laser radar on the vehicle, and finally accurately determine the position and posture of the point cloud, so that the data can be converted into the earth coordinate system.
[0033] Specifically, the laser radar is installed on the roof and maintained in a horizontal position. The laser radar has the advantages of long measurement distance, high scanning frequency, high ranging sampling rate and high measurement resolution. The higher scanning frequency can ensure that the laser radar installed on a fast-moving car can still obtain a clear three-dimensional point cloud. However, when the sampling frequency is fixed, a faster scanning frequency will reduce the angular resolution, so generally only a moderate scanning frequency can be selected.
[0034] The laser radar collects point cloud data frame by frame with a collection angle of 360 degrees. The field of view is composed of an elliptical cone formed by the horizontal angle α and the vertical angle β between the driver and the target point (the center of the cone).
[0035] The inertial navigation system includes a positioning and a directional antenna. The inertial navigation system can generate 150 measurements per second, provide the geodetic coordinates of each measurement of the vehicle, and obtain the trajectory of the vehicle. The laser radar collects point clouds 20 times per second. Each point cloud collected uses the laser radar as the coordinate origin, but it is not point cloud data in the geodetic coordinate system. By synchronizing the time and space of the laser radar and the inertial navigation system, the point cloud data can be converted to the geodetic coordinate system.
[0036] The specific evaluation method includes the following steps:
[0037] S1 obtains vehicle trajectory and point cloud data through lidar and inertial navigation system, and accurately determines the position and posture of the point cloud based on the vehicle trajectory.
[0038] S2 pre-processes and separates the point cloud data to form ground point cloud data and point cloud data above the ground;
[0039] Specifically:
[0040] Due to the measurement principle, the resolution of LiDAR will drop dramatically as the distance of the detected object increases. The point cloud data should be preprocessed first, and the point cloud data close to the scanning center should be retained, that is, the area where the point cloud data is more concentrated. This area should contain most of the point cloud, and the points contained in it should have the characteristics of clear imaging, small error, and high density, so as to eliminate the distant and messy point clouds in the LiDAR point cloud.
[0041] The retained point cloud data is separated using the RANSAC segmentation algorithm to form ground point cloud data and point cloud data above the ground.
[0042] S3 uses the Delaunay triangulation algorithm to form a TIN triangulation network for ground point cloud data, and uses the TIN triangulation network to extract the linear data of the road. It first divides the road into segments according to the combination of road lines, and extracts and calculates the three-dimensional features of the road: unit tangent vector θ, principal normal vector β, secondary normal vector γ, three-dimensional curvature k, and three-dimensional torsion τ, and finally forms a three-dimensional geometric feature set of the road.
[0043]
[0044]
[0045]
[0046]
[0047]
[0048] Furthermore, the Delaunay triangulation consists of a set of connected but non-overlapping triangles, and the circumscribed circles of these triangles do not contain any other points in the region.
[0049] Furthermore, the road alignment combination mainly refers to the alignment combination of the road plane and longitudinal section. The plane alignment is divided into straight lines, circular curves, and transition curves, and the longitudinal section is divided into longitudinal slope sections and vertical curve sections.
[0050] Using the existing three-dimensional running speed prediction model, the three-dimensional characteristics of five types of roads are substituted to obtain the running speed of vehicles on the road.
[0051] The three-dimensional running speed prediction model can refer to the paper: Research on the prediction method of passenger car running speed based on the linear three-dimensional spatial geometric characteristics of highway.
[0052] For different commonly used running speeds, linear interpolation can be performed in combination with the value distribution of the horizontal maximum viewing angle 4 and the vertical maximum viewing angle 5 of the dynamic viewing angle to determine the range of the cone ellipse formed by the corresponding viewing angle under the running speed.
[0053] S4 uses a classification algorithm to classify and reconstruct the point cloud data above the ground to form a defined three-dimensional object, i.e., a homogeneous object. Other point cloud data generates multi-patch files to form heterogeneous objects. Irrelevant point cloud parameters that do not belong to road information are eliminated, such as other vehicles, pedestrians, etc. Only a three-dimensional environment containing road information can be seen. The road information includes the edge lines on both sides of the road and obstacle information that affects the road cone, such as guardrails.
[0054] S5 takes the edge line on one side of the one-way road as the most unfavorable position of the visibility, and the edge line on the other side as the most favorable position of the visibility. The smaller of the two is the three-dimensional dynamic available visibility of the driver in a real three-dimensional environment.
[0055] The three-dimensional dynamic sight distance is that in the three-dimensional line shape of the highway, when the vehicle is traveling at a certain speed, the driver searches along the extension direction of the road within the dynamic viewing angle, starting from the first point on the near side lane edge line that can be seen until the first point that cannot be seen. The distance from this point to the observation point along the route is the three-dimensional dynamic sight distance of the observation point.
[0056] The specification can determine the offset position of the driver's viewpoint relative to the center of the vehicle under different standard vehicle models. Generally, the default position where the target object can be seen is 0.1m high on the edge line of the near side lane.
[0057] The driver's line of sight and the tangent of the lane line remain parallel and consistent.
[0058] The line of sight relationship between the ellipse formed by the driver's viewpoint in the dynamic viewing angle and the target object point is derived through the principle of line of sight between two points in space. If the target object point is blocked by obstacles such as the road itself and objects formed by surrounding facilities, it is not line of sight, otherwise it is in a line of sight state.
[0059] Starting from the lane edge line corresponding to the center point of the vehicle, the intervals between the target points are all 5m, forming a visual cone of the driver's viewpoint and the target point and moving forward to the first target point that is not visible. The distance between the previous target point of the target point and the driver's route is the three-dimensional dynamic available sight distance of the vehicle at this position. The viewpoint positions of different drivers are calculated in turn along the vehicle's forward direction at intervals of 5m to determine the available sight distance of the entire road section.
[0060] Specifically: on a road containing lane dividing line 10 and road marking line 12 information, when the driver of the vehicle is in the most unfavorable sight distance vehicle position 9, the vehicle is close to the lane edge line 3 of the road, and the road side obstacle area 8, 13 and the road line shape combined occlusion established by point cloud data are used. The driver's viewpoint 1 is combined with the maximum horizontal dynamic angle 4 and the maximum longitudinal dynamic angle 5 at the running speed to form a cone of view with the viewpoint center line 6 as the center position, and the position 7 of the farthest target object point on the lane center line 2 and the long distance of the curve projected from the driver's viewpoint 1 to the lane edge line 11 along the tangent direction of the road can be seen as the three-dimensional available sight distance.
[0061] Assuming that the left and right lane edge lines of the vehicle deviating from the one-way road are the most unfavorable and most favorable positions of the sight distance, the three-dimensional dynamic sight distance under the two positions is obtained, and the smaller value is the three-dimensional dynamic available sight distance of the driver in a real three-dimensional environment.
[0062] The driving sight distance of the vehicle at that position is calculated according to the formula provided in the specification, and compared with the calculated three-dimensional dynamic available sight distance at that position. If the three-dimensional dynamic available sight distance is greater than the required driving sight distance, the sight distance verification requirements are met, otherwise they are not met.
[0063] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.
Claims
1. A three-dimensional dynamic line of sight evaluation method based on lidar point cloud data, It is characterized in that The steps include: Obtain vehicle trajectory and point cloud data; Preprocess and separate the point cloud data to form ground point cloud data and point cloud data above the ground; The Delaunay triangulation algorithm is used to form a TIN triangulation network for the ground point cloud data, and the linear data of the road is extracted using the TIN triangulation network; the road is segmented according to the combination of linear shapes to obtain the three-dimensional geometric feature set of the road; the running speed prediction model based on the three-dimensional geometric features is used to obtain the running speed of the vehicle, and the dynamic viewing angle range of the driver at each moment is determined according to the running speed; Classify the point cloud data above the ground to form homogeneous objects and non-homogeneous objects, and further obtain a three-dimensional environment containing only road information; The most unfavorable position of the sight distance is when the vehicle is biased toward the edge line of one side of the one-way road, and the most favorable position of the sight distance is when the vehicle is biased toward the other side. The smaller sight distance between the two is the 3D dynamic available sight distance of the driver in a real 3D environment. The most unfavorable position of the sight distance is when the vehicle is biased toward the edge line of one side of the one-way road, and the most favorable position of the sight distance is when the vehicle is biased toward the other side. The smaller sight distance between the two is the three-dimensional dynamic available sight distance of the driver in a real three-dimensional environment, which is: When the driver is in the most unfavorable sight distance vehicle position, the farthest position of the lane edge line that the driver can see is used as the three-dimensional dynamic sight distance by using the cone ellipse surface and the arc length of the route corresponding to the current position of the lane edge line of the driver; When the driver is in the most favorable sight distance vehicle position, the farthest position of the lane edge line that the driver can see is used as the three-dimensional dynamic sight distance by using the cone ellipse surface and the arc length of the route corresponding to the current position of the lane edge line of the driver; The 3D dynamic viewing distances obtained from the two positions are compared, and the smaller value is the 3D dynamic available viewing distance of the driver in a real 3D environment; The dynamic viewing angle is the maximum horizontal and vertical viewing angle that the driver can see in the dynamic field of view state. The elliptical cone of each driver's position is determined by this viewing angle, and its projection to the vertical two-dimensional space is the cone elliptical surface.
2. The three-dimensional dynamic viewing distance evaluation method according to claim 1, It is characterized in that The preprocessed point cloud data is processed by the RANSAC segmentation algorithm to form ground point cloud data and point cloud data above the ground.
3. A system for implementing the three-dimensional dynamic viewing distance assessment method according to any one of claims 1 to 2, It is characterized in that It includes a laser radar and an inertial navigation system, wherein the inertial navigation system includes an inertial navigation unit and a GPS unit.
4. The system according to claim 3, It is characterized in that The laser radar is arranged on the top of the vehicle and keeps a horizontal position.
5. The system according to claim 3, It is characterized in that The inertial navigation unit is installed on the center line of the rear wheel of the vehicle with the Y-axis direction facing the forward direction of the vehicle.
6. A system according to any one of claims 3 to 5, It is characterized in that The timestamps of the LiDAR and INS are synchronized, and the origin of the point cloud collection coordinate system starts at the LiDAR position, and the INS provides the position of the vehicle in the Earth coordinate system.
7. The system according to claim 3, It is characterized in that Intra-frame correction and feature extraction are performed using the LOAM algorithm.
Citation Information
Patent Citations
Method for dynamically measuring sight distance of drivers based on video
CN102230794A
Expressway sight distance detection method based on laser radar
CN109613544A
Method and apparatus for evaluating sight distance
US20080021680A1
