Research For Visual Tracking Algorithm Under Varying Illumination

In recent years, with the continuously development of computer hardware and sensor technology, visual tracking has become a very popular research topic in computer visual and pattern recognition fields and can be applied in civilian and military fields. Although many effective visual tracking methods have been proposed, there are still some factors resulting in lower accuracy when identifying moving targets and extracting their features, such as occlusion and clutter. There are many factors changing features of moving targets. For example, color feature of target surface will change when illuminating varing and edge feature will change when the posture of target varing. There are many factors changing gray and content of adjacent images, such as slight camera shake and shadow, etc. The research of visual features extracting and robust of visual tracking algorithm in complex environments has a profound theoretical significance and broad practical value. This paper makes deep study on visual tracking algorithm in complex environment.Firstly, it analyzes feature selection and extraction methods of traditional vision tracking. In view of its poor self-adapting to illumination, the paper proposes a feature extraction algorithm self-adapting to illumination on the basis of color feature extraction algorithm, which is of good comprehensive properties. Unstable track problem caused by using unique kind of features can be settled by building fusion feature set through dynamic selecting color feature and edge feature, etc. Fisher Criterion is used to evaluate feature identifying ability and renew feature set online, which enhances adaptive ability of fusion features. Fuzzy control method is used to contrl renew and then enhance renewing stability. The adaptability of visual feature renewing stability of fusion feature are improved by using our method.Secondly, it analyzes calibration methods and PTZ adjusting algorithm in visual tracking system. To settle the problem of poor initiative and poor stability caused by acquisition environment with varying illumination, it put forwards an active PTZ adjusting method based on Fuzzy control. Target pre-location method based on local particle filter using illumination invariants is used to realize automatic calibration estimation for moving targets, which realizes PTZ pre-adjusting and improves the initiative of system adjustment. Filter sampling method using illumination invariants overcomes the effects of illumination changes and noises etc to target pre-location and strengthen system robustness. Fuzzy control method is used to control Pan and Tilt to improve the stability of visual tracking system. Experiment proves that our method can improve initizative and accuracy of PTZ adjustment to some extent.Thirdly, after analyzes tracking algorithm for moving target of visual tracking system, it selects color feature-based tracking algorithm with better performance as a base. To settle the problem of illumination sensibility, it presents a visual tracking algorithm robust to varying illumination, which reduces illumination sensitivity by selecting self-adapting fusion feature as base. It makes use of target shadow characteristics to realize object segment and detection, which enhances the accuracy of recognition and tracking for object under shadow. Tracking method using particle filtering can realize nonlinear, non-Gaussian and multi-state motion tracking. To some extent, the algorithm improves tracking accuracy and illumination adaptability.Robust multi-targets tracking method based on improved Mumford-Shah model is proposed to settle the problem of sensitive to varying illumination of tracking algorithm and solve the difficult of tracking for multi-target with occlusion using single camera after analyzing multi-targets tracking algorithm. It selects edge feafure and target model as basis to realize identifying to multi-targets in complex environments using improved Mumford-Shah model which reduce the effect of blurred edges and noise. Then, It utilizes method of pixel marked to record the target and background regions. In tracking sence with fixed target number, Independent and complete edge-feature model can be built up even when the targets are occluded each other. A joint particle filter algorithm is put up to realize multi-target tracking separately. The methods can bring down computational complexity and reduce the influences of illumination changes and blur edges. Compared with traditional difference algorithm or tracking algorithm based on color feature, the method is of better robustness and accuracy.Finally, visual tracking experimental platform is designed to realize test and analysis of the algorithm in complex environment. It designs a menu function to select parameters so as to obtain experimental results quickly and objectively. In addition, it has a comparative function of algorithm performance which is benefit to make an analysis of testing results and enhance experimental effectively. It also provides an extentsible software platform in order to carry out further research work.The research is carried out around visual tracking algorithms in variable illumination environments, in which it involves the effects of varying illumination to visual tracking system in every important link. It can improve the overall performance of system, which provides a better theoretical basis for future work.