Vision target tracking method aiming at target scale change

Abstract

The invention relates to a vision target tracking method in the field of computer vision, and particularly relates to a vision target tracking method aiming at target scale change. The method comprises the following steps: 1) defining affine transformation of the target in a geometric particle filtering frame and in an Aff (2) space, and establishing a first-order autoregression motion model of the affine transformation; and 2) expressing an observation model of the target in a linear approximation mode through vector characteristics of a candidate area by adopting the vector characteristics of a template, wherein coefficients are subjected to non-negative constraint. The obtained optimization problem is just the non-negative least square problem, and the similarity of a candidate with a target template is reflected by the expressed coefficients, so that the coefficient can be used for defining the observation likelihood of the candidate target. The final tracking result is obtained by calculating sample mean values defined on the Aff (2) space. With the adoption of the method, the target when having scale or rotation change can be tracked better.

Description

technical field [0001] The invention relates to a visual target tracking method in the field of computer vision. Specifically, it involves visual object tracking methods for object scale changes. Background technique [0002] Video object tracking is to estimate the position of visual objects in image sequences, and it has always been a hot topic in computational vision. It is often necessary to identify and track the target with the help of the features such as the edge, color and texture of the target itself. In addition to the determination of the target position, the estimation of the scale size is also very important. The relationship between them is like that between the egg and the chicken. If the target scale estimation is too large or too small, it will affect the accuracy of target positioning. Conversely, the estimated deviation of the position will also affect the scale estimation. [0003] Visual object tracking methods are roughly divided into two categorie...

AI Technical Summary

Problems solved by technology

The disadvantages are: 1) The algorithm itself does not have a window adaptive mechanism, and needs to use external methods to obtain the target scale

2) Easy to get stuck in the local minimum point of the target, 3) Cannot recover from the loss of the target

[0008] Text [R.Liu, Z.Jing, H.Pan, A novel algorithm based on nonnegative least-square estimation for visual object tracking, Optical Engineering, 51(3), pp.037201-1-037201-7, 2012.] It is based on the particle filter of the vector space, and the complex motion is not considered enough

Images