CN-115685249-B - Obstacle detection method and device, electronic equipment and storage medium

CN115685249BCN 115685249 BCN115685249 BCN 115685249BCN-115685249-B

Abstract

The invention discloses a method, a device, electronic equipment and a storage medium for detecting an obstacle, which comprise the steps of acquiring multi-frame initial point clouds acquired by a multi-line laser radar and pose data acquired by an IMU (inertial measurement unit) when a vehicle runs, and carrying out de-distortion and fusion on the multi-frame initial point clouds based on the pose data to obtain target point clouds; inputting the target point cloud into a preset obstacle identification model to obtain azimuth information and size information of a small obstacle, wherein the small obstacle is an obstacle with a height within a preset height range. The multi-frame initial point cloud is subjected to de-distortion processing based on pose data, so that accuracy of point cloud data can be improved, the problem that small obstacles are difficult to detect due to point cloud distortion is avoided, the multi-frame correction point cloud is fused into the target point cloud, the point cloud density of the target point cloud is improved, the point cloud density of the small obstacles in the target point cloud is improved, characteristics of the small obstacles can be expressed more completely, and detection of the small obstacles is facilitated.

Inventors

ZHOU QIAN
LAI ZHILIN
LI LIANGYUAN
LUO ZENGHUI
WANG XIN

Assignees

广州赛特智能科技有限公司

Dates

Publication Date: 20260512
Application Date: 20221107

Claims (9)

1. An obstacle detection method, comprising: when a vehicle runs, acquiring multi-frame initial point clouds acquired by a multi-line laser radar and pose data acquired by an IMU; Performing de-distortion processing on a plurality of frames of initial point clouds based on the pose data to obtain a plurality of frames of corrected point clouds; Fusing a plurality of frames of correction point clouds into a target point cloud; Inputting the target point cloud into a preset obstacle identification model to obtain azimuth information and size information of a small obstacle, wherein the small obstacle is an obstacle with a height within a preset height range; the preset obstacle recognition model is trained by the following modes: Acquiring a training point cloud set comprising a plurality of point cloud subsets, wherein when the point cloud subsets are point cloud subsets of a preset type, the point cloud subsets comprise point clouds corresponding to small obstacles; Labeling each preset type point cloud subset with first tag information and first confidence coefficient, wherein the first tag information comprises first position information and first size information of the small obstacle, the first confidence coefficient is the confidence coefficient of the first tag information, Initializing an obstacle recognition model; randomly extracting the point cloud subset and inputting the point cloud subset into the obstacle recognition model to obtain second tag information and second confidence coefficient of the point cloud subset, wherein the second tag information comprises second azimuth information and second size information of the small obstacle; Calculating an error by using the first tag information, the first confidence coefficient, the second tag information and the second confidence coefficient; judging whether the error is smaller than a preset error threshold value or not; If yes, stopping training the obstacle recognition model to obtain a trained obstacle recognition model; If not, adopting the error to adjust the model parameters of the obstacle recognition model, and returning to the step of randomly extracting the point cloud subset and inputting the point cloud subset into the obstacle recognition model; Labeling the first label information for each point cloud subset, including: Generating a radar image according to the point cloud subsets for each point cloud subset; Marking a minimum cube surrounding a small obstacle in the radar image; and taking the azimuth information and the size information of the minimum cube as first tag information, wherein the azimuth information comprises the central coordinate and the direction angle of the minimum cube.
2. The method of claim 1, wherein said de-distorting the initial point cloud of multiple frames based on the pose data to obtain a corrected point cloud of multiple frames, comprising: determining a reference point from the initial point cloud for each frame, and taking other points except the reference point in the initial point cloud as distortion points; performing motion compensation on points in the initial point cloud by adopting the pose data to obtain interpolation pose data of each point; Calculating the spatial pose relation of the distortion point and the datum point according to interpolation pose data between the time stamp of the distortion point and the time stamp of the datum point for each distortion point; And in the initial point cloud of each frame, converting the distortion point into a coordinate system of the datum point according to the space pose relation corresponding to the distortion point, and obtaining a correction point cloud.
3. The method of claim 2, wherein the reference point is a last point in the initial point cloud, and the performing motion compensation on the points in the initial point cloud by using the pose data to obtain interpolation pose data of each point comprises: for each point in the initial point cloud, taking the pose data of the front frame and the rear frame of the timestamp corresponding to the point as reference pose data; And taking the motion variable quantity of the two frames of the reference pose data as interpolation pose data of points.
4. The method of claim 1, wherein the fusing the plurality of frames of the correction point clouds into a target point cloud comprises: Determining a first point cloud and a plurality of frames of second point clouds acquired before the first point cloud from a plurality of frames of correction point clouds; Converting the multiple frames of second point clouds into point clouds under a coordinate system where the first point clouds are located, and obtaining multiple frames of third point clouds; and fusing a plurality of frames of the third point cloud and the first point cloud into a target point cloud.
5. The method of claim 4, wherein converting the plurality of frames of the second point cloud into a point cloud in a coordinate system in which the first point cloud is located, to obtain a plurality of frames of a third point cloud, comprises: Acquiring a first conversion matrix from a coordinate system of each frame of the second point cloud to a map coordinate system of the vehicle and a second conversion matrix from the map coordinate system to the coordinate system of the first point cloud; Converting a plurality of frames of second point clouds into point clouds under the map coordinate system based on the first conversion matrix to obtain a plurality of frames of fourth point clouds; And converting the multiple frames of fourth point clouds into point clouds under the coordinate system of the first point clouds based on the second conversion matrix to obtain multiple frames of third point clouds.
6. The method according to any one of claims 1-5, wherein inputting the target point cloud into a preset obstacle recognition model to obtain azimuth information and size information of a small obstacle, comprises: Inputting the target point cloud into a preset obstacle recognition model to obtain label information and confidence coefficient, wherein the label information is azimuth information and size information of a small obstacle, and the confidence coefficient is the confidence coefficient of the label information; and when the confidence coefficient is larger than a preset confidence coefficient threshold value, the label information is used as azimuth information and size information of the small obstacle.
7. An obstacle detection device for performing the obstacle detection method of any one of claims 1-6, comprising: the data acquisition module is used for acquiring multi-frame initial point clouds acquired by the multi-line laser radar and pose data acquired by the IMU when the vehicle runs; the de-distortion module is used for performing de-distortion processing on the multi-frame initial point cloud based on the pose data to obtain multi-frame correction point cloud; the point cloud fusion module is used for fusing a plurality of frames of correction point clouds into a target point cloud; and the target information extraction module inputs the target point cloud into a preset obstacle recognition model to obtain azimuth information and size information of a small obstacle, wherein the small obstacle is an obstacle with a height within a preset height range.
8. An electronic device, the electronic device comprising: at least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the obstacle detection method of any one of claims 1-6.
9. A computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, the computer instructions for causing a processor to perform the obstacle detection method of any one of claims 1-6.

Description

Obstacle detection method and device, electronic equipment and storage medium Technical Field The present invention relates to the field of autopilot technology, and in particular, to a method and apparatus for detecting an obstacle, an electronic device, and a storage medium. Background In the field of automatic driving, environmental awareness is a prerequisite for an automatic driving vehicle to safely and reliably perform tasks, and an obstacle detection function is a basic function in an environmental awareness system. Lidar is a primary sensor for detecting obstacles, whose effect is generally related to the stability of radar detection. The MEMS micro-vibrating mirror scanning scheme adopted by the solid-state radar has the advantages that the required laser transmitters and receivers are very few, only the micro-vibrating mirror swings during working, the stability of detecting obstacles is high, but the cost is high, the technical maturity is lower than that of the multi-line laser radar, and therefore in the prior art, the object detection is mainly carried out through the multi-line laser radar. The adjacent wire harnesses of the multi-wire-harness laser radar point cloud are sparse, so that for a large obstacle, the collected points are relatively more, the effect of detecting the large obstacle is better, but for a small obstacle, the collected points are relatively less, and then the detection system may not be capable of determining the point cloud corresponding to the small obstacle from the collected point clouds, so that the detection effect of the small obstacle is unstable. Disclosure of Invention The invention provides an obstacle detection method, an obstacle detection device, electronic equipment and a storage medium, which are used for solving the problem that the detection effect of a small obstacle is unstable when object detection is carried out by using a laser radar point cloud with a low wire harness at present. In a first aspect, the present invention provides a method for detecting an obstacle, including: when a vehicle runs, acquiring multi-frame initial point clouds acquired by a multi-line laser radar and pose data acquired by an IMU; Performing de-distortion processing on a plurality of frames of initial point clouds based on the pose data to obtain a plurality of frames of corrected point clouds; Fusing a plurality of frames of correction point clouds into a target point cloud; Inputting the target point cloud into a preset obstacle recognition model to obtain azimuth information and size information of a small obstacle, wherein the small obstacle is an obstacle with a height within a preset height range. In a second aspect, the present invention provides an obstacle detection device comprising: the data acquisition module is used for acquiring multi-frame initial point clouds acquired by the multi-line laser radar and pose data acquired by the IMU when the vehicle runs; the de-distortion module is used for performing de-distortion processing on the multi-frame initial point cloud based on the pose data to obtain multi-frame correction point cloud; the point cloud fusion module is used for fusing a plurality of frames of correction point clouds into a target point cloud; the target information extraction module is used for inputting the target point cloud into a preset obstacle recognition model to obtain azimuth information and size information of a small obstacle, wherein the small obstacle is an obstacle with a height within a preset height range. In a third aspect, the present invention provides an electronic device, including: at least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the obstacle detection method according to the first aspect of the invention. In a fourth aspect, the present invention provides a computer readable storage medium storing computer instructions for causing a processor to execute the obstacle detection method according to the first aspect of the invention. The obstacle detection method comprises the steps of obtaining multi-frame initial point clouds collected by a multi-line laser radar and pose data collected by an IMU when a vehicle runs, performing de-distortion processing on the multi-frame initial point clouds based on the pose data to obtain multi-frame correction point clouds, fusing the multi-frame correction point clouds to obtain target point clouds, inputting the target point clouds into a preset obstacle recognition model to obtain azimuth information and size information of small obstacles, wherein the small obstacles are obstacles with heights within a preset height range. The multi-frame initial point cloud is subjected to de-distortion processing based on pose data, the accuracy of the point cloud data can be improved, the point cloud data can reflect env