3D target tracking method and system based on point cloud sequence data

Abstract

The invention discloses a 3D target tracking method and system based on point cloud sequence data, and belongs to the field of digital image recognition. The 3D target tracking method comprises the following steps: respectively extracting and standardizing a search point cloud and a template point cloud containing a target frame from a current frame and a previous frame, and predicting the position and posture of the target frame in the current frame by utilizing a 3D target tracking model so as to determine the position of a 3D target in the current frame, wherein in the 3D target tracking model, the feature extraction network is used for extracting template point cloud features and searching point cloud features, and the correlation prediction network is used for predicting a target score of each feature point in the search point cloud, and the integrated regression network is used for carrying out point-by-point regression after the two features are fused, and the position prediction network is used for carrying out weighted multiplication on the distance and posture between each feature point in the fused features and the center of a target box according to the target score ofeach feature point of the search point cloud. According to the 3D target tracking method, the three-dimensional attribute of the object can be fully utilized, and the calculation efficiency and stability of 3D target tracking are improved.

Description

technical field [0001] The invention belongs to the field of digital image recognition, and more specifically relates to a 3D target tracking method and system based on point cloud sequence data. Background technique [0002] In the fields of autonomous driving and robot movement trajectory planning, it is usually necessary to perceive and understand the surrounding environment, detect the trajectory of the surrounding target objects, and then plan its own movement direction. Among them, detecting the target trajectory requires the use of target tracking technology. The traditional 2D object tracking technology has achieved remarkable results, but it can only track the movement of the object on the camera plane, so it cannot meet the requirements of automatic driving and robot trajectory planning that need to obtain the three-dimensional motion information of the object. [0003] Compared with 2D target tracking technology, 3D target tracking technology can make good use of ...

AI Technical Summary

Problems solved by technology

Among them, the 3D target tracking method based on RGBD video sequences is very dependent on 2D target tracking technology, which uses depth map information for further processing on the basis of 2D target tracking results, and does not make full use of 3D object structure information; The 3D target tracking method performs overhead projection on the 3D point cloud data, and then uses the 2D target tracking method to track, which loses the target information in the height direction, and cannot effectively track the target with a large change in the height direction.

[0004] Compared with the 3D object tracking method based on RGBD video sequence and the 3D object tracking method based on bird's-eye view video sequence, the 3D object tracking method based on point cloud sequence data can make full use of the three-dimensional properties of the object, but the point cloud data has no The existing 3D target tracking method based on point cloud sequence data has a search space when predicting the position of the target frame in the current frame point cloud data based on the tracking results of the previous frame point cloud data. It is extremely large, so the calculation efficiency is low, and the prediction model cannot be trained end-to-end, and the stability of the tracking results cannot be guaranteed

Images