Deform able object tracking algorithm based on visual salience and system thereof

Abstract

The invention discloses a Deform able object tracking algorithm based on visual salience and a system thereof. The method includes 1, calibrating a target region manually; 2, by means of an image segmentation algorithm based on the visual salience, acquiring a partial salience spectrum containing a prospect target; 3, segmenting the prospect region; 4, establishing a target prospect model; 5, initializing the prospect model through a color histogram tracker on the basis of color features; 6, acquiring and judging an alternate target through an optical flow tracker on the basis of point features; 7, acquiring the final position of the target; 8, learning and updating the target. According to the method, the salience processing is performed, the algorithm calculating efficiency is considered, the prospect information is segmented automatically by the partial size window method, a softening model is established through prospect information matrix, a color histogram corresponding to the model is generated for histogram matching, the judging capability of the tracker is improved, and the novel model of the target is learned continuously through a learner.

Description

technical field [0001] The invention relates to a visual saliency-based tracking algorithm of deformable objects and a system thereof. Background technique [0002] In recent years, with the rapid development of machine learning and pattern recognition related fields, the development of computer vision technology has been greatly promoted, and a large number of engineering practice and algorithm papers have sprung up like mushrooms. CVPR, ECCV, and ICC, the core journals in the field of computer vision, have a large number of highly innovative and influential excellent papers. Among them, online learning, as a machine learning method, has been more and more introduced into the field of computer vision, and a series of online learning targets can be developed, and the template model can be continuously updated, so as to achieve faster and more robust Algorithmic technology for tracking. [0003] Computer Vision The target tracking problem of computer vision based on online ...

AI Technical Summary

Problems solved by technology

[0008] The purpose of the present invention is to overcome the deficiencies of the prior art, provide a deformable object tracking algorithm and system based on visual salience, and solve the problem of using a rigid template to model the target in the prior art, which often deforms and has an ugly appearance. Regular objects are modeled with a rectangular frame template with mismatched shapes, which will inevitably lead to wrong learning of background pixels, making it very easy to be disturbed by background changes during the tracking process. If the target itself is also deformed at the same time, it is extremely Problems that can easily lead to tracking failures

Images