Discriminative online target tracking method based on videos in dictionary learning

Abstract

The invention belongs to the field of image processing technologies and relates to a discriminative online target tracking method based on videos in dictionary learning. The method includes the first step of constructing a template of a target through manual calibration or according to an existing tracking result and obtaining the state of the target to be tracked at the same time, the second step of constructing an initial stage matrix of the target, conducting motion modeling and conducting sampling to form candidate samples, the third step of partitioning the candidate samples and the template, conducting sparse coding on all partitions and obtaining expression coefficients of the corresponding partitions, the fourth step of inputting the expression coefficients as features into a classifier of the corresponding partition, obtaining the determination confidence degree of each candidate sample in the partition, randomly selecting K determination confidence degrees of the partitions, conducting summation, traversing all possibilities and selecting the candidate sample with the maximum summation value as the target tracking result of a current frame, wherein the selected frequency of the candidate sample in all the possibilities are maximum. Through the method, the accuracy and robustness of target tracking on the condition of a complicated background and changing of target appearance can be improved.

Description

technical field [0001] The invention relates to a method in the technical field of image processing, in particular to a method for discriminative online target tracking in video based on dictionary learning. Background technique [0002] In recent years, video surveillance technology has been widely used. Target tracking integrates knowledge and technologies in many related fields such as computer image processing, pattern recognition, artificial intelligence, and automatic control. Hierarchy analysis and understanding provide important data basis. Stable and accurate tracking of targets in complex environments is an urgent problem to be solved in the current research and application of tracking technology, and traditional tracking methods often perform poorly. In recent years, a class of methods that can adaptively update the relevant parameters of the tracker as the target and the environment change, called online target tracking, has received extensive attention. This ...

AI Technical Summary

Problems solved by technology

[0009] The present invention aims at the above-mentioned deficiencies existing in the prior art, and proposes a discriminative online target tracking method in video based on dictionary learning, which changes the problem of poor performance of traditional tracking methods in complex backgrounds, and is suitable for complex scenes (severe motion, illumination changes, noise, partial occlusion and attitude changes, etc.) to achieve robust target tracking

Images