Target tracking method based on manifold discriminant non-negative matrix factorization

Abstract

The invention discloses a target tracking method based on manifold discriminant non-negative matrix factorization. The target tracking method comprises the following steps: S1: obtaining a positive sample and a negative sample of a current frame; S2: obtaining the characteristics of the positive sample and the negative sample and a sample matrix X1; S3: reading a next frame, and obtaining a candidate sample matrix Xu; S4: combining X1 with Xu as a data matrix X, decomposing X into a non-negative matrix product, and learning to obtain a classifier; S5: through the classifier, calculating the response value of each candidate sample, and selecting a maximum response as a tracking target; and S6: judging whether the current frame is a last frame or not, entering S7 if the current frame is the last frame to output the state of each frame of target, and otherwise, jumping to S1. By use of the target tracking method, through the non-negative matrix factorization, higher-level image features are obtained, local characteristics can be better described, and shielding and background interference can be eliminated. A semi-supervised manifold regularization method is used and is combined with marked and unmarked samples to train the classer which contains spatial structure information, more discriminant information can be retained, and illumination and target deformation can be effectively coped with. A feature extraction model is trained and updated on line to quickly position an appointed target in a video.

Description

technical field [0001] The invention relates to a target tracking method based on manifold discrimination non-negative matrix decomposition, and belongs to the technical fields of computer vision, pattern recognition and video monitoring. Background technique [0002] Object tracking aims at locating, identifying a specified object in a scene from a video, and estimating its trajectory. Many high-level tasks in computer vision, such as scene understanding, event detection, action recognition, etc., rely heavily on the results of object tracking. In the top international academic journals and conferences in the field of computer vision and pattern recognition, object tracking occupies a considerable amount of space and proportion. As a cutting-edge research direction that integrates computer vision, image processing, pattern recognition, machine learning, statistical analysis and stochastic processes, target tracking is widely used in video surveillance, human-computer inter...

AI Technical Summary

Problems solved by technology

[0008] The purpose of the present invention is to solve the lack of discriminative ability of the appearance model or the trained classifier used in the existing target tracking method, which cannot effectively deal with the problems of occlusion, target deformation and displacement, and illumination changes in complex scenes, which leads to the easy loss of targets and tracking failures, a target tracking method based on manifold discriminant non-negative matrix factorization is proposed

Images