A target tracking method based on feature fusion

Abstract

The paper presented a adaptive template update method of multi feature fusion based on Camshift algorithm. When one feature to distinguish the foreground and background is not obvious, other features can be complement each other. The algorithm combines texture, edge and color features, using the different characteristics contribution degree set the different weight calculated in the multi feature space. It can better solve the problems of background color similar, objective morphological changes and the change of illumination. Introduction At present, the research of video object tracking technology has become the core issue in the video monitoring field. Color feature is not affected by the shape of the object changes, it has the scale and rotation invariant characteristics, the amount of calculation is small, the existing algorithms mostly use the features to represent the target. But in complex traffic scene with the uneven brightness, high noise, close background color[1,2], it's easy to be effected by the disruptor and cause tracking failure. According to the complex traffic environment, Camshift fused LTP texture model, HSV model and edge features, made feature contribution degree to obtain the feature weight. Experiments show that, the new algorithm search fast, low complexity, there is a better robustness to traffic high noise, nonuniform illumination, color similar complex environment. Camshift Algorithm Meanshift algorithm is the core of Camshift, make the target object color distribution is u q , moving target in the Ith frame color distribution is ( ) u p i y ∧ . Use Bhattacharrya coefficient to judge the similarity of color distribution. Let 1 i y + be the next frame motion target center and seek 1 i y + let a, b most similar. 1 1 u 1 ( ) p ( )q m