Optical Flow Estimation Based On Feature Points Moving Target Tracking Algorithm Theory And Analysis

It is an extremely challenging subject in computer vision to track the inter-ested object effectively in image sequence. In recent years, the theory of motion target tracking has been widely applied in some important fields, such as military guidance, security monitoring, biomedical science, video coding, etc. Tracking al-gorithms mushroom quickly through the unremitting efforts of scientific researcher around the world. However, the existing methods are proposals for the certain given environment or some target condition. The diversity of the target's characteristics and the complexity of the environment determine that the tracking technology are facing the great difficulties in practical application. This paper has done some re-search on motion target tracking in the status of stationary background.Several corner operators based on grayscale are analyzed and compared on the aspects of algorithm's idea and operating procedure. Each operator is evaluated in performance on the basis of simulation experiment. Furthermore, the classical Harris detection algorithm is optimized.This paper describes two algorithms of motion estimation method, which are motion estimation method based on gradient and motion estimation method based on correlation. An improved method is purposed for optimizing the Horn-Schunck algorithm. New algorithm doesn't add the smoothing constraint condition on the whole image any longer. For reducing the impact on motion boundary, the smoothing constraint condition is pulsed on the direction which is perpendicular to the gradient.This paper presents a feature-point matching algorithm based on optical flow es-timation, which simultaneously compensates for the shortage of optical flow method and corner matching method. It is a high-precision motion estimation algorithm. The experimental results indicate that this algorithm is more applicable to the situations of the partially covered object, rotation, and noise disturbance. It can achieve the matching correctly and reduce the matching time. For the motion object tracking, this algorithm has some theoretical significance and practical value.