Sparsely represented selective appearance model-based frame-adaptive target tracking algorithm

Abstract

The invention discloses a sparsely represented selective appearance model-based frame-adaptive target tracking algorithm. The algorithm comprises the steps of randomly sampling a plurality of candidate targets by an initialization tracker; running a discrimination tracker SDC; conducting the frame characteristics detection; running a generative tracker SGM; updating an occlusion detection threshold Th; calculating a joint model and determining to track. Compared with the prior art, on the foundation that the defect in the prior art that the learning-based super-resolution reconstruction algorithm is dependent on a large number of training sets is sovled, the neighborhood embedding method is improved. Meanwhile, the problem that the high-frequency initial estimation of the super-resolution algorithm, based on local self-similarity and multi-scale similarity, is not accurate is also solved. The super-resolution reconstruction effect of images is improved. The results of experiments show that, the aliasing effect and the ringing effect are better suppressed based on the above algorithm. Meanwhile, high-resolution images that are closer to real images are reconstructed. Therefore, the algorithm is better in subjective quality and objective quality.

Description

technical field [0001] The present invention relates to the field of object tracking of computer vision, and more particularly, relates to a frame-adaptive object tracking algorithm using a sparse representation selective appearance model. Background technique [0002] As a cutting-edge subject of computer vision, object tracking technology is a research hotspot in both scientific and engineering fields. The task of target tracking is to obtain the position information of the target of interest from each image frame, including coordinates, size, rotation angle and even speed information, to realize the analysis and understanding of the object's trajectory, so as to achieve more advanced functions. There are many problems to be solved in target tracking, such as target occlusion, rotation, scale change, brightness change, etc. Therefore, the appearance representation method of tracking target is an important research topic in this field. [0003] Sparse representation is a r...

AI Technical Summary

Problems solved by technology

However, this method is not adaptive to different frames, resulting in high computational complexity, and the use of image information is not accurate enough.

[0004] So far, no frame-adaptive target tracking algorithm based on sparse representation selective appearance model has been developed in the papers and documents published at home and abroad. Therefore, the invention content of this patent is original

Images