The Study Of HCI System Based On Moving Object Trajectory Recognition

Moving object trajectory recognition based on computer vision is a key technology in human computer interaction area since it can give an alarm in good time for the abnormal situation by recognizing the moving object trajectory in intelligent monitor. Moreover, it can also provide a more natural and harmonious way for HCI. The real-time moving object trajectory recognition based on a single camera in complex background is an important development direction. Moving object trajectory recognition based on computer vision can be divided into four stages:moving object detection, moving object tracking, trajectory feature extraction and feature recognition. Through consulting great deal of literature and datum, the contributions of this thesis are mainly focused on the following three respects in order to build a trajectory recognition system.In the moving object detection stage, this thesis proposed an improved algorithm, referred to as motion aim division algorithm, which combines the adaptive Gaussian mixture background model and object color detection based on HSV color space.In the moving object tracking stage, this thesis proposed an improved algorithm based on traditional Camshift tracking algorithm. The binary image obtained from the background subtraction of improved GMM background model do AND operation with the color probability distribution image in Camshift algorithm.The treated color probability distribution image can effectively remove the interference of the small amount of similar color background. This paper introduces the Kalman filter to estimate the target location in the image sequence, effectively improving the ability of anti-interference of the system. To enhance the robustness of the system, this paper also proposed an adaptive extended search window algorithm, allowing the system to instantly track the object again after missing it.In the trajectory feature extraction stage, this paper presents a rapid and practical method of feature extraction for this system. In the trajectory feature recognition stage, discrete hidden Markov model, which is from left to right, is applied to model, train and recognize the trajectory feature. Experiments show that the system can better recognize the moving object trajectory, and can achieve the requirements of accurate and real-time.