An improved motion pedestrian tracking algorithm based on CamShift

Target detection, recognition and tracking are particularly important in intelligent driving. With the development of artificial neural network research, the recognition and tracking algorithm based on neural network has been greatly improved in recognition speed and accuracy. But its performance depends greatly on the training database, and the amount of calculation is too large to meet the real-time requirements. The tracker based on correlation filtering is fast, but it will lose the target once the target is slightly occluded. In this paper, we proposed an improved algorithm based on CamShift and Kalman filter. Through the prediction function of Kalman filter, reduce the range of search window of CamShift. Then the target information obtained by CamShift algorithm is fed back to Kalman filter for updating and correction. The optimization algorithm not only satisfies the real-time requirement of video target tracking, but also improves the tracking accuracy even if the target is overshadowed.