Kalman Predicting Based Target Tracking Of Man-machine Interface

Computer vision is a very new research field which has developed very rapidly. It has also become an important interdisciplinary field of study. Moving targets real-time detection, recognition and tracking in complex background is an important research task in Automatic Target Recognition (ATR) field. It is widely used in security monitoring, virtual reality, human-computer interaction and other fields. At the same time, there are high academic value and theoretical significance in moving target detection and tracking research field. And they have become one of the international research hotspot in recent years.This paper presented a tracking algorithm based on Camshift and Kalman prediction which was proposed to solve the poor tracking ability problem in occlusions just using single Camshift. Firstly, Inter-frame difference threshold method was used to achieve detection and extraction of the target rapidly and accurately. Secondly,we used the color characteristics of moving objects in Camshift algorithm to find its location and size in image sequences .Finally, to deal with the object occlusion by other objects which have similar color,a Kalman filter was used to track the object region centroid.We used wireless remote car to achieve a moving target tracking experiment. The experimental results denoted that combining CamShift algorithm and Kalman prediction filtering could complete a real-time and accurate target tracking.In this paper we achieved a novel method of man-machine interface. Firstly, we detected and focused on the face area using Adaboost algorithm.Then a co]or histogram model was constructed after the more precise area of the face was got.Secondly,Camshift algorithm estimated the center and size of human face tracking window.To increase the tracking speed ,a two-dimensional Kalman filter was used to track the object region centroid(human face).Finally,the three degrees of freedom of human face were used to control over the role of computer games in six sports.The experimental results showed that this method was robust and efficient.