Vehicle separation method based on single-line laser radar

Abstract

A vehicle separation method based on a single-line laser radar comprises the following steps: (1) calibrating the sensor of a single-line laser radar vehicle separator installed for the first time; (2) after obtaining the angle, obtaining the coordinates of the point cloud of each frame of data in the horizontal coordinate system according to the angular resolution; (3) obtaining the number N of ground point clouds and the clustering height H of the point clouds when no vehicle passes through the ETC, and using the number N of ground point clouds and the clustering height H of the point clouds as coarse filtering parameters to determine whether the frame has vehicle data; after (3), dividing the frame of point clouds into intervals according to the x-axis direction, equally dividing them into S intervals, obtaining the intervals with the largest and second largest number of point clouds, obtaining the interval K where the outermost point cloud of the vehicle is located, and determining whether any of the intervals with the largest and second largest number of point clouds per frame is K. If not, it is determined that there is no vehicle, and if it is K, it is initially determined that there is a vehicle. The present invention can simply and effectively separate vehicles passing through the ETC lane, with low cost and high recognition.

Description

Technical Field

[0001] This patent application belongs to the field of ETC, and specifically relates to a technology for separating vehicles passing through an ETC lane. Background Art

[0002] In recent years, with the improvement of transportation infrastructure, the continuous increase in the mileage of highways and the number of bridges, people's travel has become more and more convenient. But at the same time, the traffic flow at various highway intersections, bridges, tunnels and other toll booths has also increased significantly. It is particularly important to improve the efficiency of toll collection, and vehicle separation and vehicle model information detection are important supports for the toll collection system. When a vehicle passes through EAC, the vehicle and other objects are separated by a vehicle separator. Then, it is necessary to prepare separation technology to achieve accurate identification.

[0003] CN210091375U discloses a scanning laser vehicle separation device based on FPGA. The patent includes a housing and a rotating reflection module installed in the housing, a pulse laser emission and receiving module, an FPGA and a power module, and a data output module. The FPGA processes the single-turn ranging data by angle segment to detect whether there is a ranging point of the vehicle scanned in each data segment. If the data segment contains a vehicle ranging point, the IO output is triggered to achieve the purpose of rapid vehicle detection. The single-turn data of the vehicle is integrated and transmitted to the calculation and processing module to construct a three-dimensional point cloud of the vehicle and extract vehicle feature information from it. However, the time-scanning laser vehicle separator used in the patent has many disadvantages such as inaccurate separation of obstacle point clouds and ground point clouds, long time consumption, and low real-time performance.

[0004] CN108427124A discloses a multi-line laser radar ground point separation method, device and vehicle, which adopts a multi-line laser radar. For obstacle detection of the multi-line laser radar, obstacles can be accurately identified, and obstacle (i.e. foreground) point cloud and ground (background) point cloud can be accurately separated. However, it adopts a multi-line laser radar, which makes the algorithm more complicated and the cost higher. Summary of the invention

[0005] The technical problem to be solved by the present invention is: how to separate vehicles passing through ETC, therefore, a vehicle separation method based on a single-line laser radar is provided.

[0006] The technical solution adopted by the present invention is:

[0007] A vehicle separation method based on a single-line laser radar comprises the following steps:

[0008] (1) Calibrate the sensor of the single-line laser radar vehicle separator that has been installed for the first time: Calculate the angle α between the ground plane and the detection horizontal plane through ground point cloud fitting;

[0009] (2) After obtaining the angle α, the coordinates of the point cloud of each frame of data in the horizontal coordinate system can be obtained according to the angular resolution:

[0010]

[0011]

[0012] (3) The number of ground point clouds N and the clustering height H of the point clouds when no vehicle passes through the ETC are obtained, and the number of ground point clouds N and the clustering height H of the point clouds are used as coarse filtering parameters to determine whether there is vehicle data in the frame; the clustering height H is the average height in the horizontal coordinate system; let {M p} is the point set after being filtered by threshold A in the horizontal coordinate system:

[0013]

[0014] (4) For the point cloud judged as yes after filtering in (3), divide the frame point cloud into intervals along the x-axis direction, and divide it equally into S intervals. Calculate the intervals with the largest and second largest number of point clouds, and obtain the interval K where the outermost point cloud of the vehicle is located. Determine whether any of the intervals with the largest and second largest number of point clouds in each frame is K. If not, it is judged that there is no vehicle. If it is K, it is initially judged that there is a vehicle.

[0015] The method further includes step (5) of determining the point cloud distribution characteristics of the front and rear frame data: determining whether the point cloud in the frame is a vehicle or a noise point based on the similarity of the point cloud distribution on the outside of the vehicle when the vehicle passes.

[0016] Step 1 includes:

[0017] (1.1) Collect a continuous point cloud when no vehicle passes through ETC {N p}, assume that M frames of point cloud data are collected, and the initial zero point is close to the horizontal upward. First, by setting the meta data threshold A, the meta data threshold A is the dis sampling distance. In actual operation, the dis sampling distance is combined with the sensor detection distance and the ETC lane width setting. The noise is filtered by setting the meta data threshold A to obtain the actual data point cloud {N' p}

[0018] {N′ p}={N p |dis meta <A}

[0019] (1.2) According to the meta distance dis meta , angular resolution θ and frame index i, the coordinates of each frame data point cloud are obtained

[0020]

[0021]

[0022] (1.3) Set the height threshold B to obtain the bottom point cloud combination {D p}

[0023]

[0024] (1.4) Combining point clouds {D p}Perform least squares plane fitting to obtain the fitting plane, calculate the plane inclination and the angle α between the ground plane and the detection horizontal plane required for calibration.

[0025] According to the fitting plane, the intersection line of the fitting plane and the horizontal plane is obtained, and then perpendicular lines are drawn from the fitting plane and the horizontal plane to the intersection line respectively, and the angle between the two perpendicular lines is obtained, so as to obtain the angle α between the ground plane and the detected horizontal plane.

[0026] Step (4) comprises:

[0027] (4.1) Divide the point cloud into S intervals according to equal distances, and record the interval index as 1 to S;

[0028] (4.2) Accumulate the point clouds of consecutive P frames, and continuously record the interval indexes with the largest and second largest number of point clouds in each frame, which are recorded as the interval index K1 with the largest number of point clouds and the interval index K2 with the second largest number of point clouds, respectively;

[0029] (4.3) Count the number of occurrences of the interval index K1 with the largest number of point clouds and the interval index K2 with the second largest number of point clouds in the P frame, and assign the value with the largest number of occurrences as K;

[0030] (4.4) Judge the point cloud of each frame of the continuous P frames. If the point cloud with the largest number or the second largest number is K, it is initially judged to be a car, otherwise it is not a car;

[0031] (4.5) Starting from the P+1 frame data, repeatedly calculate the interval index K1 with the largest number of point clouds and the interval index K2 with the second largest number of point clouds, update K, and repeat the above steps to determine whether it is a car.

[0032] Step (5) comprises:

[0033] (5.1) Obtain the KL distance between the current frame L and the previous and next R frames, and form a KL distance combination {KL 2R};

[0034] (5.2) For the KL distance combination {KL 2R}Analyze: If the number of adjacent L frames whose KL distances are less than the threshold C satisfies D, it is determined to be a car, otherwise there is no car.

[0035] The present invention adopts the above-mentioned technical solution, which can simply and effectively separate the passing vehicles in the ETC lane while relying only on a single-line laser radar, and has the characteristics of low cost, high recognition and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a coordinate system relationship diagram of the present invention. DETAILED DESCRIPTION

[0037] A vehicle separation method based on a single-line laser radar comprises the following steps:

[0038] (1) Perform sensor calibration on the newly installed single-line laser radar vehicle separator: Calculate the angle α between the ground plane and the detection horizontal plane through ground point cloud fitting. Since the sensor calibration only needs to be calculated once, it needs to be performed during the debugging phase after the vehicle separator is installed. The main contents are as follows:

[0039] (1.1) Collect a continuous point cloud when no vehicle passes through ETC {N p}, assume that M frames of point cloud data are collected, and the initial zero point is close to the horizontal upward. First, by setting the meta data threshold A, the meta data threshold A is the dis sampling distance. In actual operation, the dis sampling distance is combined with the sensor detection distance and the ETC lane width setting. The noise is filtered by setting the meta data threshold A to obtain the actual data point cloud {N' p}

[0040] {N′ p}={N p |dis meta <A}

[0041] (1.2) According to the meta distance dis meta , angular resolution θ and frame index i, the coordinates of each frame data point cloud are obtained

[0042]

[0043]

[0044] (1.3) Set the height threshold B (which can be obtained by experiment) and find the bottom point cloud combination {D p}

[0045]

[0046] (1.4) Combining point clouds {Dp} Perform the least squares plane fitting, get the fitting plane, then get the intersection line of the fitting plane and the horizontal plane, then draw perpendicular lines from the fitting plane and the horizontal plane to the intersection line, and find the angle between the two perpendicular lines, then get the plane inclination and the angle α between the ground plane and the detection horizontal plane required for calibration. The coordinate relationship involved is as follows: Figure 1 shown.

[0047] (2) After obtaining the angle α, the coordinates of the point cloud of each frame of data in the horizontal coordinate system can be obtained according to the angular resolution:

[0048]

[0049]

[0050] (3) Calculate the number of ground point clouds N and the cluster height H of the point clouds when no vehicle passes through the ETC, and use the number of ground point clouds N and the cluster height H of the point clouds as coarse filtering parameters to determine whether there is vehicle data in the frame. The cluster height H is the average height in the horizontal coordinate system. Let {M p} is the point set after being filtered by threshold A in the horizontal coordinate system:

[0051]

[0052] (4) After filtering in (3), the point cloud judged as yes is divided into intervals along the x-axis direction. It can be equally divided into S intervals, and the intervals with the largest and second largest number of point clouds are obtained. This method is to obtain the interval K where the outermost point cloud of the vehicle is located. According to whether any of the intervals with the largest and second largest number of point clouds in each frame is K, if it is not K, it is judged that there is no car. If it is K, it is initially judged that there is a car. The steps are as follows:

[0053] (4.1) Divide the point cloud into S intervals according to equal distances, and record the interval index as 1 to S;

[0054] (4.2) Accumulate the point clouds of consecutive P frames, and continuously record the interval indexes with the largest and second largest number of point clouds in each frame, which are recorded as the interval index K1 with the largest number of point clouds and the interval index K2 with the second largest number of point clouds, respectively;

[0055] (4.3) Count the number of occurrences of the interval index K1 with the largest number of point clouds and the interval index K2 with the second largest number of point clouds in the P frame, and assign the value with the largest number of occurrences as K;

[0056] (4.4) Judge the point cloud of each frame of the continuous P frames. If the point cloud with the largest number or the second largest number is K, it is initially judged to be a car, otherwise it is not a car;

[0057] (4.5) Starting from the P+1 frame data, repeatedly calculate the interval index K1 with the largest number of point clouds and the interval index K2 with the second largest number of point clouds, update K, and repeat the above steps to determine whether it is a car.

[0058] (5) In order to further determine the point cloud initially identified as a vehicle and distinguish it from noise such as exhaust gas, it is necessary to identify the distribution characteristics of the point cloud of the previous and next frame data. Based on the similarity of the point cloud distribution on the outside of the vehicle when the vehicle passes, it can be determined whether the point cloud in the frame is a vehicle or a noise point. The specific steps are as follows:

[0059] (5.1) Obtain the KL distance between the current frame L and the previous and next R frames, and form a KL distance combination {KL 2R}, that is, the KL distances between the L frame and the LR, L-R+1, L-R+2…L, L+1, L+2…L+R frames;

[0060] (5.2) For the KL distance combination {KL 2R} for analysis, if the number of adjacent L frames whose KL distance is less than the threshold C satisfies D, it is determined to be a car, otherwise it is considered to be no car. (The determination of C / D is set based on experimental experience).

[0061] It should be noted that the process of obtaining the KL distance between two discrete distributions is:

[0062] For two discrete distributions p(x) and q(x), the KL divergence distance is defined as:

[0063]

[0064] D KL The smaller (p||q), the greater the similarity between p(x) and q(x), and vice versa.

Claims

1. A vehicle separation method based on a single-line laser radar, characterized in that: The steps include: (1) Calibrate the sensor of the single-line laser radar vehicle separator that has been installed for the first time: Calculate the angle α between the ground plane and the detection horizontal plane through ground point cloud fitting; (2) After obtaining the angle α, the coordinates of the point cloud of each frame of data in the horizontal coordinate system can be obtained according to the angular resolution: Among them, dis meta is the meta distance; the angular resolution θ and the frame index i; (3) The number of ground point clouds N and the clustering height H of the point clouds when no vehicle passes through the ETC are obtained, and the number of ground point clouds N and the clustering height H of the point clouds are used as coarse filtering parameters to determine whether there is vehicle data in the frame; the clustering height H is the average height in the horizontal coordinate system; let {M p } is the point set after being filtered by threshold A in the horizontal coordinate system: (4) For the point cloud judged as yes after filtering in (3), divide the frame point cloud into intervals along the x-axis direction, and divide it equally into S intervals. Calculate the intervals with the largest and second largest number of point clouds, and obtain the interval K where the outermost point cloud of the vehicle is located. Determine whether any of the intervals with the largest and second largest number of point clouds in each frame is K. If not, it is judged that there is no vehicle. If it is K, it is initially judged that there is a vehicle.

2. The vehicle separation method based on single-line laser radar according to claim 1 is characterized in that: The method further includes step (5) of determining the point cloud distribution characteristics of the front and rear frame data: determining whether the point cloud in the frame is a vehicle or a noise point based on the similarity of the point cloud distribution on the outside of the vehicle when the vehicle passes.

3. The vehicle separation method based on single-line laser radar according to claim 1 is characterized in that: Step 1 includes: (1.1) Collect a continuous point cloud when no vehicle passes through ETC {N p }, suppose that M frames of point cloud data are collected, and the initial zero point is close to the horizontal upward. First, by setting the meta data threshold A, the meta data threshold A is the dis sampling distance. In actual operation, the dis sampling distance is combined with the sensor detection distance and the ETC lane width setting. The noise is filtered by the set meta data threshold A to obtain the actual data point cloud {N′ p } {N′ p }={N p |say meta <A} (1.2) According to the meta distance dis meta , angular resolution θ and frame index i, the coordinates of each frame data point cloud are obtained (1.3) Set the height threshold B to obtain the bottom point cloud combination {D p } (1.4) Combining point clouds {D p }Perform least squares plane fitting to obtain the fitting plane, calculate the plane inclination and the angle α between the ground plane and the detection horizontal plane required for calibration.

4. The vehicle separation method based on single-line laser radar according to claim 3 is characterized in that: According to the fitting plane, the intersection line of the fitting plane and the horizontal plane is obtained, and then perpendicular lines are drawn from the fitting plane and the horizontal plane to the intersection line respectively, and the angle between the two perpendicular lines is obtained, so as to obtain the angle α between the ground plane and the detected horizontal plane.

5. The vehicle separation method based on single-line laser radar according to claim 1 is characterized in that: Step (4) comprises: (4.1) Divide the point cloud into S intervals according to equal distances, and record the interval index as 1 to S; (4.2) Accumulate the point clouds of consecutive P frames, and continuously record the interval indexes with the largest and second largest number of point clouds in each frame, which are recorded as the interval index K1 with the largest number of point clouds and the interval index K2 with the second largest number of point clouds, respectively; (4.3) Count the number of occurrences of the interval index K1 with the largest number of point clouds and the interval index K2 with the second largest number of point clouds in the P frame, and assign the value with the largest number of occurrences as K; (4.4) Judge the point cloud of each frame of the continuous P frames. If the point cloud with the largest number or the second largest number is K, it is initially judged to be a car, otherwise it is not a car; (4.5) Starting from the P+1 frame data, repeatedly calculate the interval index K1 with the largest number of point clouds and the interval index K2 with the second largest number of point clouds, update K, and repeat the above steps to determine whether it is a car.

6. The vehicle separation method based on single-line laser radar according to claim 1, characterized in that: Step (5) comprises: (5.1) Obtain the KL distance between the current frame L and the previous and next R frames, and form a KL distance combination {KL 2R }; (5.2) For the KL distance combination {KL 2R }Analyze: If the number of adjacent L frames whose KL distances are less than the threshold C satisfies D, it is determined to be a car, otherwise there is no car.

Images