Video target tracking state analysis method, device and implementation device

Abstract

The invention provides a video target tracking state analysis method and device and an implementation device. The method comprises the steps that an initial feature point set of a target in a currentvideo frame is acquired; the initial feature point set comprises multiple feature points; according to the distance between each feature point and the central position of the target, the feature points classified by mistake are detected; the feature points classified by mistake are eliminated, and a first feature point set is generated; the position and range of the target are determined accordingto the first feature point set, and apparent features of the target are extracted within the position and range; and according to the apparent features, whether the target in the current video frameis subjected to tracking drift is analyzed. Through the method, the change of the tracking state in the video target tracking process can be analyzed comprehensively and accurately without the need for priori knowledge of a target motion track, good adaptability is achieved, and therefore the steadiness and robustness of target tracking are improved.

Description

technical field [0001] The invention relates to the technical field of video object tracking, in particular to a video object tracking state analysis method, device and implementation device. Background technique [0002] The purpose of video target tracking is to accurately locate the position and range of the target in each frame of video. However, affected by factors such as noise, cluttered background, and occlusion, drift, target occlusion, and even tracking loss will inevitably occur during the tracking process. Condition. To achieve robust and robust tracking, it is necessary to accurately understand whether the tracking of the current frame is accurate, whether tracking drift, occlusion or even tracking loss occurs, and adjust the tracking strategy accordingly according to the analysis of the current tracking results to achieve the purpose of robust tracking . Existing target tracking state analysis methods are usually based on trajectory analysis or eigenvector an...

AI Technical Summary

Problems solved by technology

Existing target tracking state analysis methods are usually based on trajectory analysis or eigenvector analysis. Trajectory-based tracking state analysis methods can evaluate the tracking state more accurately, but require prior knowledge of the target's motion constraints, which limits its The scope of application, the analysis method based on eigenvectors does not need to have constraints on the movement of the target, but it is difficult to accurately distinguish between tracking drift and even tracking loss. Generally speaking, the existing methods for analyzing the target tracking state are vulnerable to related parameter settings , threshold selection, noise and clutter, the analysis results are often not reliable

Images