Target tracking method based on supervised significance detection

Abstract

The invention discloses a target tracking method based on supervised significance detection. The target tracking method comprises steps that a searching area of a current frame is divided into super pixels, and super pixel characteristics of a target and a background are extracted, and a support vector machine SVM is used to learn the discriminant appearance model; each time when a new frame of image occurs, the super pixel segmentation of the searching area is carried out, and first-stage significance detection is carried out by using manifold sequencing based on a graph model; the probability of every super pixel of the new frame of image belonging to the target is calculated according to the discriminant appearance model, and classification results are adjusted, and by combining with the first-stage significance detection, a classification result is adjusted, and random walk seed points are selected by combining with the first-stage significance detection, and a second-stage saliency map is acquired by adopting random walk; by adopting the weighting of the saliency map and the classification result, a confidence graph is acquired, and by processing the confidence graph, an integral image is used to estimate the new position and the new dimension of the target. Problems such as rapid motion and deformation are effectively processed, and therefore robustness tracking is realized.

Description

technical field [0001] The present invention relates to the field of computer vision, and more specifically, to a method for object tracking based on supervised saliency detection. Background technique [0002] As an important research direction in the field of computer vision, object tracking has received extensive attention. This technology has broad application prospects in the fields of security monitoring, unmanned driving and military defense. Although there are already a considerable number of object tracking methods, these methods are often unstable or even invalid under the conditions of illumination changes, object deformation, fast motion and severe occlusion. Therefore, proposing an effective target tracking algorithm has important application value and practical significance. [0003] Target tracking has developed rapidly in recent years, and effective target modeling is extremely important for tracking. In order to design robust appearance models, a visual r...

AI Technical Summary

Problems solved by technology

[0004] On the other hand, the appearance model is an important part of the tracking problem. Many discriminative models based on boosting, MIL, and SVM have been continuously developed. However, these methods mostly use a rectangular box to represent the target, and usually use a global appearance model. Although this can Handle a certain degree of local deformation, which is not suitable for tracking some non-rigid bodies that undergo severe deformation

Images