Sparse tracking method on basis of gradient texture features

Abstract

The invention provides a sparse tracking method. Gradient texture features and affine transformation parameters of targets are jointly used for building appearance models of the targets. The sparse tracking method includes steps of creating initial dictionaries; extracting candidate particles; creating the gradient texture features; extracting gradient texture features of dictionary templates; extracting gradient texture features of candidate samples; linearly expressing the candidate samples by all templates in the dictionaries; solving sparse coefficients; creating reconstruction errors. The sparse tracking method on the basis of the gradient texture features has the advantages that the problem of change of the appearance of targets due to rotation, scale and illumination change and the like can be effectively solved by the aid of the gradient texture features; challenge due to scale variation or view angle variation can be effectively handled by means of affine transformation; the templates can be updated by the aid of an incremental sub-space learning process, accordingly, tracking drifting can be effectively suppressed, and problems of target occlusion and the like can be solved; the target tracking stability can be kept in complicated scenes, and the precision of trackers can be improved.

Description

technical field [0001] The invention belongs to the field of image processing and computer vision, and relates to a tracking method based on gradient texture features. It can be applied to fields such as human-computer interaction and video surveillance. Background technique [0002] Object tracking occupies an extremely important position in the field of computer vision, such as automatic surveillance, video retrieval, and human-computer interaction. At present, many researchers have made some achievements in theory and application for the problem of target tracking. [0003] Mei et al. (Xue Mei, Ling Haibin. Robust visual tracking using l1minimization[C]. Computer Vision, 2009IEEE 12th International Conference on. IEEE, 2009: 1436-1443.) took the lead in introducing sparse representation to the tracking problem and achieved better results. The algorithm combines the target template and the occlusion template to form a dictionary, and then uses the linear combination of a...

AI Technical Summary

Problems solved by technology

However, many important features of the image, such as edges and corner features, are easily affected by curvature flow or local curvature flow, resulting in partial curve edges becoming blurred under the action of curvature flow.

Images