Target tracking method and device

Abstract

The invention belongs to the technical field of image processing, and particularly relates to a target tracking method, which comprises the following steps: obtaining a video to be tracked, determining and inputting an initial frame from the video to be tracked, and initializing a target frame; generating a plurality of candidate boxes in the next frame after the first frame, wherein the candidateboxes are the same as the area in the first frame; extracting features of candidate boxes in the same area image; screening and extracting background features and target features in the features, andperforming preliminary detection on the target features; modeling the screened background features and target features, respectively scoring the features in the candidate boxes, calculating confidence scores, and determining the highest confidence as a prediction result. The method has the advantages that the characteristics of the shielded target can be detected and correspondingly processed, the robustness of the algorithm is enhanced, the tracking stability is ensured, and the problem that the tracking effect becomes poor due to large-area shielding in the target tracking process is solved.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a target tracking method and device. Background technique [0002] Object tracking is one of the hotspots in the field of computer vision research, and has been widely used in many fields. Generally speaking, target tracking is to establish the positional relationship of the object to be tracked in a continuous video sequence, and obtain the complete trajectory of the object. In simple terms, target tracking is to give the initial position of the target in the first frame, and use the tracking algorithm to calculate the position information of the target in each subsequent frame of image. In theory, target tracking can perform real-time tracking, but in practical applications In , due to factors such as illumination, occlusion, and scale changes, it is easy to cause target loss. [0003] The information loss and complex application environment in the existing i...

AI Technical Summary

Problems solved by technology

In simple terms, target tracking is to give the initial position of the target in the first frame, and use the tracking algorithm to calculate the position information of the target in each subsequent frame of image. In theory, target tracking can perform real-time tracking, but in practical applications However, due to factors such as illumination, occlusion, and scale changes, it is easy to cause the target to be lost.

[0003] The information loss and complex application environment in the existing image signal acquisition process, especially the target is often occluded by other objects in the scene during the movement process, the occlusion leads to the loss of the original features and the generation of new features, which makes the target recognition difficult. and positioning pose great difficulties

Therefore, the occlusion problem in the tracking process has increasingly become a key factor limiting the practicability of tracking algorithms, and traditional recognition methods in image pattern recognition are usually ineffective for such object recognition problems

Images