CN113903179B - Using method of multi-beam laser radar background filtering device based on point cloud density superposition distribution - Google Patents

Abstract

The invention relates to a method for using a multi-line beam laser radar background filtering device based on the superimposed distribution of point cloud density, and belongs to the technical field of road traffic monitoring. First, a period of data frames are aggregated based on the coordinates of the lidar points, and then a three-dimensional matrix is established to represent the entire space. The matrix elements are cubes. The density of the aggregated points in the cube is recorded, and the point density threshold of the cube is determined. Cubes higher than the threshold are background cubes. , the ones below the threshold are non-background cubes. After identifying the background cubes that determine the threshold, save the background cubes into a three-dimensional matrix, and use the three-dimensional matrix as the background matrix. The background matrix is combined with real-time data, and the points collected in the real-time data can be displayed in the background. If it is found in the matrix, it will be excluded, and if it is not found in the background matrix, it will be retained to achieve background filtering.

Description

A method of using a multi-beam lidar background filtering device based on the superimposed distribution of point cloud density

technical field

The invention relates to a method for using a multi-line beam laser radar background filtering device based on the superimposed distribution of point cloud density, and belongs to the technical field of road traffic monitoring.

Background technique

Connected vehicle technology can connect all road users, sharing real-time location, speed and direction, thereby helping to avoid accidents, saving time costs and reducing fuel consumption and pollution emissions. However, the number of road users who install IoV devices is limited, and unconnected users and connected users will exist for a long time. In mixed traffic situations where unconnected and connected users coexist, IoV technology can only obtain partial vehicle motion information, and a new method is needed to collect high-resolution microscopic flow data to fill in the data caused by mixed traffic flow vacancy.

In recent years, advanced self-driving cars have begun to employ lidar sensors to detect road boundaries and lane markings. The in-vehicle system obtains the necessary data from the point cloud output of the lidar sensor to determine potential obstacles present in the environment and the vehicle's relationship to those potential obstacles. As an important sensor of the Internet of Vehicles technology, LiDAR can perform 360° high-precision scanning of surrounding target objects, and at the same time track and report the precise position and speed of each target object within its scanning range. Installing it on the roadside can fill mixed traffic. The data gap caused by traffic provides an effective solution.

To obtain high-resolution micro-flow data of all road users, background filtering is a preprocessing step and the basis for improving data accuracy and computational efficiency. Background filtering refers to culling all objects except road targets. The so-called background includes static objects such as trees, buildings and traffic facilities, as well as dynamic objects such as swaying branches and noise points. In previous studies, many methods have been developed for background filtering of different data types. For example, an algorithm that uses per-frame pixel information to extract a background image from a traffic video stream analyzes the color value of each pixel in a series of frames and then uses the pattern of that sequence as the correct color value for the background image. In addition, there is a way to calculate the difference between background and non-background using the median value of each color channel. The results show that these methods are well suited for highways in free-flow to moderately congested conditions. But these vision-based data processing algorithms cannot be used directly for lidar data, because roadside lidar data is a series of points, not raster pixel information. Therefore, existing background filtering methods cannot directly extract data from roadside lidar.

SUMMARY OF THE INVENTION

In view of the deficiencies of the prior art, the present invention provides a method for using a multi-line beam lidar background filtering device based on the superimposed distribution of point cloud density, which is applied to different driving environments such as roads, bridges and tunnels, and is used after scanning by lidar. The background filtering method is processed to obtain high-resolution microscopic flow data.

Terminology Explanation:

Aggregation: refers to the content selection, analysis, and classification of relevant data, and finally analyzes to obtain the desired results. The aggregation of the present invention refers to the analysis and integration of the collected point cloud data frames, and multiple data frames are formed into a single data. The frame is transmitted, thereby increasing the transmission content and improving the efficiency for subsequent background recognition.

The technical scheme of the present invention is as follows:

A method for using a multi-beam lidar background filtering device based on the superimposed distribution of point cloud density, the device includes a lidar and a control terminal, the lidar is connected to the control terminal, data collection is performed through lidar scanning, and the control terminal is used for laser detection. The data collected by the radar is processed, and the steps are as follows:

(1) When applied to the middle section of the road and the bridge, first erect the support rod, set the laser radar on the support rod, and connect the laser radar to the control terminal;

(2) 360° scanning of lidar for point cloud data collection, collecting raw data within a period of time as initial input;

(3) Aggregate data frames for a period of time based on the coordinates of lidar points; the more frames, the higher the accuracy, the increased time cost and the higher requirements for computer memory.

(4) Cut the three-dimensional space into continuous cubes:

A three-dimensional matrix is established to represent the entire space (the entire space refers to the data space obtained after the data obtained by the lidar is aggregated), the matrix elements are cubes, and the gathering points in the cube are recorded (the gathering points are the focused 3D points, including background points and non-background points. )quantity;

(5) Determine the point density threshold of the cube through the density of the aggregation points in each cube to distinguish the background cube from the non-background cube, and the density is the ratio of the number of aggregation points in each cube to the volume of the cube, which is higher than the threshold The cubes are background cubes, and those below the threshold are non-background cubes; the point density at which the lidar scans the same object varies with its distance from the lidar, generally speaking, if the distance between the target and the lidar sensor increases, the The number decreases, among which, as the distance between the vehicle and the lidar sensor increases, the value range and the number of points follow a power function relationship, so when the distance between the background object and the lidar sensor is different, the background point density is also different, which means that the threshold value It should be different in different detection ranges, usually, the density of the background space is higher than that of moving vehicles or pedestrians;

(6) After identifying the background cube for which the threshold is determined, save the background cube into a three-dimensional matrix, and use the three-dimensional matrix as the background matrix. The background matrix is combined with real-time data, and the points collected in the real-time data can be found in the background matrix and excluded. If it is not found in the background matrix, it is retained to achieve background filtering.

Preferably, moving vehicles and pedestrians generate low-density cubes. In the background frame, the number of cubes of different densities is consistent, while the variation of the number of low-density cubes for pedestrians and vehicles is consistent. Therefore, the point density threshold of the cube is determined by

Ni is the number of aggregation points of the _i -th cube, and the order of cube numbering is increasing from the least number of points to the more number of points;

F _i is the frequency of _Ni , and frequency refers to the number of cubes with the same number of points;

Slope is the slope between the aggregation point and the aggregation point;

When the slope is first 0 or positive, the frequency F of the number of points per cube in the formula is used as the threshold; when the slope is negative, it is meaningless, and the threshold is selected from when it becomes 0 or positive;

Cubes above the threshold are background cubes, and below the threshold are non-background cubes.

Preferably, in step (1), the height of the support rod is 4-6m, which meets the requirements of the scanning range and avoids human touch. The upper position of the support rod is placed at a distance of 0.5m from the top of the electric box, and the electric box is connected by screws, nuts and hoops. The bracket is connected to the support rod. The control terminal is set in the electric box. The control terminal is protected from weather damage by the electric box. The electric box is equipped with a switch door to facilitate the installation, disassembly and maintenance of the built-in equipment.

Preferably, in step (1), when there are signal lights in the middle of the road and on the bridge, the signal light pole is used as the support pole.

Preferably, in step (1), when applied to a tunnel, a fixed rod is installed at the dome, a laser radar is installed on one side of the fixed rod, an electric box is fixed on the side wall of the tunnel in the same cross-section, and a control terminal is installed in the electric box , the lidar is connected to the control terminal.

Preferably, in step (3), the number of aggregated frames of lidar for 16 beams is 1500-3000, and the number of aggregated frames of lidar for 32 beams or 64 beams can be reduced according to specific conditions to ensure real-time processing efficiency, and the peak period is to ensure real-time For data analysis, the number of aggregated frames is set to 2000.

Preferably, in step (4), the side length of the cube is 0.1 m. The key parameter for cube division is the side length of the cube, which affects the number of rows and columns of a three-dimensional matrix. Shorter side lengths increase the time cost and the dynamic background points will cross the divided cubes when moving a large distance, resulting in not capturing dynamic points well, while longer side lengths may reduce the accuracy.

The beneficial effects of the present invention are:

1. The laser radar identification device used in the present invention realizes comprehensive coverage of the two-way traffic flow of the route.

2. The installation position of the present invention is the middle line of the road, which does not affect the driving of motor vehicles and ensures traffic safety.

3. The algorithm used in the present invention can improve the recognition accuracy of vehicles and pedestrians, and significantly reduce the time cost of data processing.

4. The algorithm used in the present invention can automatically learn the cube point density threshold. Only two parameters are required: the number of frames to aggregate and the side length of the cube, both giving recommended values.

5. The algorithm used in the present invention has a small amount of calculation and can be used for real-time data processing such as vehicle monitoring and pedestrian tracking.

Description of drawings

Fig. 1 is the structural schematic diagram of the present invention on the road section without street lights and the bridge in the middle line of the road;

2 is a schematic structural diagram of the present invention on a road section with street lamps on the road center line;

3 is a schematic structural diagram of the present invention on a road section with a signal light on the road center line;

4 is a schematic structural diagram of the present invention in a tunnel;

Fig. 5 is the working state schematic diagram of the present invention;

6 is a schematic flow chart of the present invention;

Among them: 1. LiDAR; 2. Support pole; 3. Electric box; 4. Data cable; 5. Control terminal; 6. Street light pole; 7. Street light; 8. Signal light pole; 9. Signal light; 10. Signal light bracket ; 11, dome; 12, fixed rod; 13, laser; 14, pavement.

Detailed ways

The present invention will be further described below with reference to the embodiments and the accompanying drawings, but is not limited thereto.

Example 1

As shown in FIG. 1 , this embodiment provides a method for using a multi-beam lidar background filtering device based on the superimposed distribution of point cloud density. The device includes a lidar 1 and a control terminal 5. The lidar is connected to the control terminal, and the lidar is connected to the control terminal through Lidar scanning is used for data collection, which can effectively detect vehicles within a range of 100m, and process the data collected by Lidar through the control terminal. The steps are as follows:

(1) When applied to the middle section of the road and the bridge, first erect the support rod 2, set the laser radar 1 on the support rod 2, and the laser radar 1 is connected to the control terminal 5;

(2) 360° scanning of lidar for point cloud data collection, collecting raw data within a period of time as initial input;

(3) Aggregate data frames for a period of time based on the coordinates of lidar points; the more frames, the higher the accuracy, the increased time cost and the higher requirements for computer memory.

(4) Cut the three-dimensional space into continuous cubes:

A three-dimensional matrix is established to represent the entire space (the entire space refers to the data space obtained after the data obtained by the lidar is aggregated), the matrix elements are cubes, and the gathering points in the cube are recorded (the gathering points are the focused 3D points, including background points and non-background points. )quantity;

(5) Determine the point density threshold of the cube through the density of the aggregation points in each cube to distinguish the background cube from the non-background cube, and the density is the ratio of the number of aggregation points in each cube to the volume of the cube, which is higher than the threshold The cubes are background cubes, and those below the threshold are non-background cubes; the point density at which the lidar scans the same object varies with its distance from the lidar, generally speaking, if the distance between the target and the lidar sensor increases, the The number decreases, among which, as the distance between the vehicle and the lidar sensor increases, the value range and the number of points follow a power function relationship, so when the distance between the background object and the lidar sensor is different, the background point density is also different, which means that the threshold value It should be different in different detection ranges, usually, the density of the background space is higher than that of moving vehicles or pedestrians.

Moving vehicles and pedestrians generate low-density cubes. In the background frame, the number of cubes of different densities is consistent, while the variation of the number of low-density cubes for pedestrians and vehicles is consistent. Therefore, the point density threshold of the cube is determined by

Ni is the number of aggregation points of the _i -th cube, and the order of cube numbering is increasing from the least number of points to the more number of points;

F _i is the frequency of _Ni , and frequency refers to the number of cubes with the same number of points;

Slope is the slope between the aggregation point and the aggregation point;

When the slope is first 0 or positive, the frequency F of the number of points per cube in the formula is used as the threshold; when the slope is negative, it is meaningless, and the threshold is selected from when it becomes 0 or positive;

Cubes above the threshold are background cubes, and those below the threshold are non-background cubes;

After identifying the background cube that determines the threshold, save the background cube into a three-dimensional matrix, and use the three-dimensional matrix as the background matrix. The background matrix is combined with real-time data. If the points collected in the real-time data can be found in the background matrix, they are excluded. If not found, keep it to achieve background filtering.

In step (1), the height of the support rod 2 is 4-6m, which meets the requirements of the scanning range and avoids human touch. The support rod is placed 0.5m away from the top of the electric box 3, and the electric box is supported by screws, nuts and hoop brackets. The electric box 3 is provided with a control terminal 5, and the control terminal is protected from weather damage by the electric box. The electric box is equipped with a switch door to facilitate the installation, disassembly and maintenance of the built-in equipment.

In step (1), when there are signal lights 9 in the middle of the road and on the bridge, the signal light pole 8 is used as a support rod, and the signal light 9 is supported on the signal light pole 8 through the signal light bracket 10, as shown in FIG. 3 . When there are street lamps 7 in the middle of the road and on the bridges, the street lamp poles 6 are used as support rods, as shown in FIG. 2 .

In step (3), the number of aggregated frames of the 16-beam lidar is 2000, and the number of aggregated frames of the lidar of 32 beams or 64 beams can be reduced according to specific conditions to ensure real-time processing efficiency.

In step (4), the side length of the cube is 0.1m. The key parameter for cube division is the side length of the cube, which affects the number of rows and columns of a three-dimensional matrix. Shorter side lengths increase the time cost and the dynamic background points will cross the divided cubes when moving a large distance, resulting in not capturing dynamic points well, while longer side lengths may reduce the accuracy.

Example 2:

A method for using a multi-line beam lidar background filtering device based on the superimposed distribution of point cloud density, the steps are as described in Embodiment 1, the difference is that in step (1), when applied to a tunnel, in the dome 11 A fixed rod 12 is installed on the side of the fixed rod 12, a laser radar 1 is installed on one side of the fixed rod 12, and an electric box 3 is fixed on the side wall of the tunnel in the same cross-section. Show.

Example 3:

A method for using a multi-line beam lidar background filtering device based on the superimposed distribution of point cloud density, the steps are as described in Embodiment 1, the difference is that in step (3), the number of aggregated frames of lidar for 16 beams is 1500 .

Example 4:

A method for using a multi-line beam lidar background filtering device based on the superimposed distribution of point cloud density, the steps are as described in Embodiment 1, the difference is that in step (3), the number of aggregated frames of lidar for 16 beams is 3000 .

Claims (6)

1. A method of using a multi-beam lidar background filtering device based on the superimposed distribution of point cloud density, the device includes a lidar and a control terminal, the lidar is connected to the control terminal, data collection is performed by lidar scanning, and the control terminal is used for data collection. The processing of the data collected by the lidar is characterized in that the use steps are as follows: (1) When applied to the middle section of the road and the bridge, first erect the support rod, set the laser radar on the support rod, and connect the laser radar to the control terminal; (2) 360° scanning of lidar for point cloud data collection, collecting raw data within a period of time as initial input; (3) Aggregate data frames for a period of time based on lidar point coordinates; (4) Cut the three-dimensional space into continuous cubes: Establish a three-dimensional matrix to represent the entire space, the matrix elements are cubes, and record the number of aggregation points in the cube; (5) Determine the point density threshold of the cube through the density of the aggregation points in each cube to distinguish the background cube from the non-background cube, and the density is the ratio of the number of aggregation points in each cube to the volume of the cube, which is higher than the threshold The cubes of are background cubes, those below the threshold are non-background cubes; (6) After identifying the background cube for which the threshold is determined, save the background cube into a three-dimensional matrix, and use the three-dimensional matrix as the background matrix. The background matrix is combined with real-time data, and the points collected in the real-time data can be found in the background matrix and excluded. If it is not found in the background matrix, keep it to achieve background filtering; In step (5), the point density threshold of the cube is determined by the following formula

Ni is the number of aggregation points of the _i -th cube, and the order of cube numbering is increasing from the least number of points to the more number of points; F _i is the frequency of _Ni , and frequency refers to the number of cubes with the same number of points; Slope is the slope between the aggregation point and the aggregation point; When the slope is first 0 or positive, the frequency F of the number of points per cube in the formula is used as the threshold; when the slope is negative, it is meaningless, and the threshold is selected from when it becomes 0 or positive; Cubes above the threshold are background cubes, and below the threshold are non-background cubes. 2. the using method of the multi-line beam lidar background filtering device based on point cloud density superposition distribution as claimed in claim 1, is characterized in that, in step (1), the height of the support rod is 4-6m, and the support rod is on the top An electric box is placed 0.5m away from the top, and a control terminal is arranged in the electric box. 3. the using method of the multi-line beam lidar background filtering device based on the superposition distribution of point cloud density as claimed in claim 1, is characterized in that, in step (1), when there are signal lights on the middle section of the road and the bridge, use signal light poles. as a support rod. 4. the using method of the multi-line beam lidar background filtering device based on the superposition distribution of point cloud density as claimed in claim 1, is characterized in that, in step (1), when being applied in tunnel, install and fix at vault The rod, the laser radar is installed on one side of the fixed rod, and the electric box is fixed on the side wall of the tunnel in the same cross section. 5 . The method for using the multi-line beam lidar background filtering device based on the superimposed distribution of point cloud density according to claim 1 , wherein in step (3), the number of aggregated frames is 1500-3000. 6 . 6 . The method for using a multi-line-beam lidar background filtering device based on the superimposed distribution of point cloud density according to claim 1 , wherein in step (4), the side length of the cube is 0.1 m. 7 .

Images