Reserch And Implementation Of Kalman Filtering And Mean Shift In Moving Target Tracking With Adaptine Scale

Nowadays, the world is in the age of information inflating sharply. People canreceive a mass of information at everywhere and anytime. The visual information isparticularly important, so people focus on how to process the visual information moreand more. Based on the video technology analysis of moving target tracking is a hotspotof study in today’s society. Mean Shift algorithm is one of the favorite moving targettracking algorithms. To solve the target tracking window invariable and the trackingeffect is not good when the target moving fast or obscured in the traditional Mean Shift,this thesis improved the algorithms.Mean Shift algorithm using the RGB nuclear color histogram as the description oftarget model. At the same time, we extracts the geometric characteristics of the target inthe framework of Mean Shift algorithm, updating the position and size of trackingwindow according to core coordinate and target area. Finally, the Mean Shift targettemplate is updated according to what it is.To solve the fast moving or obscured target tracking problem in the Mean Shiftalgorithm and in view of Kalman filter for the prediction of moving targets, this thesiscombined Kalman filter with Mean Shift algorithm. To enhance the target locationjudgment mechanism and make more efficient target tracking, we use Kalman filter topredict the location of the target at every time of executing Mean Shift algorithm, andthen execute Mean Shift algorithm to modify the target location.This algorithm adopted by the hardware platform of DM8168multi-core processorand meanwhile using McFW framework for video processing as the processing system.The DSP core ran the target tracking algorithms. The M3VPSS is responsible for videocapture and display. The M3Video is responsible for video coding/encoding. The ARMcore is responsible for control of collaborative work between multiple cores.Based on the above research and compared with the traditional Mean Shift, weconcluded that when coping with target scale changing sharply and target movingrapidly or be obscured, the improved algorithm can obtain better tracking results.