Research Of Moving Object Detection And Tracking Based On Improved Mean-Shift Algorithm

As an important branch of computer vision, technologies of moving object detection and tracking have always been a research focus and a hot spot direction in the field of machine vision. And they are widely used in many fields such as intelligent monitoring, industrial inspection, military reconnaissance and guidance, autonomous vehicles, medical testing and so on. In this paper, we mainly study two aspects that moving object detection and tracking, improving the deficiencies of existing technical methods and innovating unceasingly. The main contents are divided into the following three points:(1) On object detection, this paper proposes a method of moving object detection based on LK optical flow algorithm and Mean-Shift algorithm. Movement of the background can bring lots of background noise and the movement of the object is irregular, so a threshold which traditional optical flow method on empirically sets to separate object and background exists inaccuracy of the large probability. Too large threshold can ignore part of the object area, but small threshold is difficult to inhibit background noise. So it leads to poor effect of object detection. According to sequence picture, a large of points of Background motion are consistent, so the optical flow motion is consistent, too. Then comparing the optical flow of all the objects with the optical flow of motiving background, total of all the objects'optical flow is different from total of all the background's optical flow and fewer. So this paper conceives directly to find out a large number of the same features of background's optical flow motion by means of gradient search principle of mean shift algorithm, regarding the same features as the threshold to separate object optical flow and background optical flow and regarding the different features as the optical flow of the object so as to accurately separate background and object, solve the problem of missing and inhabit the background noise. The way can efficiently realize the moving object detection.(2) On target tracking, based on the Mean-Shift object tracking, this paper proposes an object tracking way of many features of adaptive Mean-Shift algorithm. Aiming at the limitation of fixed nuclear window size of the Mean Shift algorithm about the change of the size of the moving object from or near the camera, this paper puts forward an automatic adjustment method of nuclear window based on edge characteristics. Use edge detection to find the center of object, obtain the two value image around the region that is slightly larger than the radius of the Mean Shift core region, and calculate Mean Shift core radius for the real target window with narrowing gradually the circular to close to the object according to the value of two images of the heart shape, which reach to adjust automatically the nuclear radius. Object model of the Mean Shift algorithm is easily affected by illumination changes, so the object model that can be updated is established in this paper with a color feature and texture feature information. When the model that can be updated is in search of the current frame, it is established around the circle with center of the current target position and the radius of the updated nuclear window. Two aspects of improvement comprehensively solve the impact of the object size and brightness changes so as to more accurate tracking.(3) Aiming at tracking problem that object is small and object moves fast, this paper combining Mean-Shift with particle filter proposes a way of tracking fast moving object. Due to exist a problem of lack of particles about particle filter, this paper first searches the approximate object's area utilizing Mean-Shift and then object is accurately tracked by the particle filter method so that the waste particles decrease and high weight particles of closing to true object's area increase which solves the problem of lack of particles of particle filter and can be efficiently applied to a tracking scene of small and fast object.