Research On Video Object Tracking With C-SIFT Agorithm

In the computer vision field, video target tracking is a meaningful research subject. Its main work is to search the most similar parts in video sequences and then it focuses on target detection, identification, extraction and tracking. According to analysis, it obtains the parameters of moving targets, such as velocity, position, acceleration, and so on. Video object tracking technology is using in many fields such as human-computer interaction, smart surveillance, intelligent buildings and national defense industry. As a meaningful research field, computer-vision-based video object tracking has attracted a large number of researchers from home and abroad, but the ideal video object tracking technology is far from mature, how to track the video object steady, quickly and timely is still a challenging subject.This paper focuses on the technology of fast video target tracking in the various environments, it presents a new tracking algorithm(C-SIFT algorithm) which is based on the image registration and coordinates locating method, and then it studies the target tracking method specifically which used the C-SIFT algorithm. The main contents of this paper and the results achieved are as follows:①This paper discusses the basic methods of the video target tracking, such as common tracking algorithms, tracking processes and technical requirement. This paper also introduces the capability requirements of the tracking algorithms. Then it discussed the advantages and disadvantages of various tracking methods.②This paper also introduces the SIFT (Scale Invariant Feature Transform) algorithm specifically, which describes the basic algorithm and nature of SIFT. Then it discusses the details of how to improve the SIFT algorithm and how to use the C-SIFT (coordinate-locating of SIFT) algorithm for target tracking.③For some feature points are not obvious in the tracking object, this paper also discusses the preprocessing methods, it focuses on the two stages of image preprocessing, the image enhancing and geometric correction theory. Experimental results show that the original image can be changed into the good matching image after preprocessing.④For C-SIFT algorithm has deficiency on the shelter in target tracking, this paper presents a new target tracking algorithm which compromises the C-SIFT algorithm and the self-adaptive Kalman filter. It uses the self-adaptive Kalman filter to estimate the initial iteration position of every frame in the tracking process and then it uses C-SIFT algorithm to calculate the tracking position. Furthermore, it adds screen ratio factor to dynamic accommodate parameter of kalman filter, then the C-SIFT algorithm can calculate the succeed state of the target. After these improvements, C-SIFT can accurately track the target even if the target was blocked for short time. The experiment results show that C-SIFT algorithm can track the object steadily and timely, and it also has a good robustness for shelter.