The Research Of Tracking Algorithm Of Moving Object Based On Video Image Sequences

Video object tracking is an integral part of computer vision and pattern recognition, which owns broad application prospects in many fields, such as visual surveillance, traffic navigation, human-computer interaction, motion analysis and so on. The challenge of visual object tracking comes from internal and external changes in the process of tracking. Internal changes reflect the changes of existing states, color and texture from tracked object. External changes mainly include the environmental changes of light, complex background and occlusion in the tracked object, which can increase the difficulty of object tracking.Firstly, aiming to the problem that it is easy to produce interference between each other according to the continuous movement of multiple objects in the video, we proposed an online multi-object tracking algorithm based on middle vision description. Through constructing the statistical model of middle vision description, establish the object scene confidence map. Adopting the principle of FIFO(First in first out) queue to train samples, aiming to the case of occluded, we proposed corresponding update mechanism, thus the problem of multi-object mutual interference can be effectively solved.Secondly, considering the traditional spatio temporal content visual tracking method only uses gray information, which ignores the color information, we proposed a simple and quick tracking algorithm. This tracking algorithm uses the main components of color and gray information as visual tracking characteristics, which can effectively deal with the colorful object. To solve the occlusion problem, firstly, we judge whether the object is occluded by the result of principal component analysis, and then the object calibration approach is proposed. Compared with the traditional object tracking algorithm, our algorithm can locate target efficiently.Finally, to connect with the unexpected pattern of video images, such as occlusion, illuminations changes and so on, we present a linear regression object tracking algorithm which based on the color properties. Extract main color in the target area of image, and then the minimum soft threshold squares distance is used to simulate the discrimination between dictionary and candidate samples. Besides, an efficient iterative method is proposed simultaneously to solve the minimum soft threshold squares distance problem, which can handle the problem of occlusion object tracking.