Target tracking algorithm based on color attributes and active feature extraction

Abstract

The invention belongs to the field of visual tracking, and provides a target tracking algorithm based on color attributes and active feature extraction for solving the problems of low speed, low robustness and inaccurate target tracking of the existing target tracking algorithm. The target tracking algorithm comprises the following steps: firstly, graying an image of a t frame, mapping the image from a RGB color space to a 11-dimensional color attribute space, mapping the image to a two-dimension by using a main component analysis method, combining gray scale information with color information, performing sampling according to a search strategy from thick to thin, mapping extracted high-dimensional multi-scale image features to a low-dimensional subspace via a very sparse random matrix by using the compressed sensing theory, finally constructing a strong classifier based on a criterion function constructed based on a Fisher information matrix, namely an active feature extraction process, classifying the features by using the strong classifier, and selecting the sample having the maximum reaction value of the classifier as the tracking target. By adoption of the method, the shortcomings of the prior art can be overcome, the algorithm obtains balance on illumination invariance and discrimination, the robustness is relatively good, the drift problem can be effectively solved, online tracking can be accomplished in real time, and the target tracking algorithm has important practical significance.

Description

technical field [0001] The invention belongs to the field of visual tracking, and relates to a compressed target tracking algorithm based on color attributes and active feature extraction. Background technique [0002] Object tracking is a very popular research topic in the field of computer vision because of its significance in vehicle navigation, traffic monitoring, and human-computer interaction. Although the subject of object tracking has been studied for decades and many tracking algorithms have been proposed, it is still a very challenging problem. Because the target appearance is disturbed by various factors, such as illumination changes, pose changes, complete or partial occlusions, and sudden movements, etc. Therefore, under the interference of the above factors, how to design a robust appearance model is a key issue in developing a high-performance tracking system. [0003] Most of the existing visual trackers rely on grayscale information or use simple color inf...

AI Technical Summary

Problems solved by technology

But it has some shortcomings: when the efficiency of the features selected by the classifier is not high, in order to make the classifier have sufficient discrimination ability, it is necessary to select relatively many features from the feature set, which will increase the time complexity of the algorithm; The more features, the greater the possibility of including low-efficiency features, and low-discrimination features will inevitably reduce the performance of the classifier and cause template drift; it does not consider the different contributions of each positive sample to the bag, resulting in the classifier extracting The possibility of inefficient positive samples increases

Images