Multi-target tracking based on Kalman filtering and optical flow histogram
Zhendi Ge, Faliang Chang, Hongbin Liu · 2017
Multi-target tracking is a challenging problem in video surveillance. The paper presents a multi-target tracking algorithm based on optical flow histogram and Kalman filtering. The method we proposed can effectively improve the tracking accuracy. First, targets are detected by Gaussion mixture model in real time. Second, a trajectory is initialized when a new target appears. Meanwhile, Kalman filtering parameters and optical flow histogram model is established. Afterwards, each trajectory associates with all detections by similarity of location and optical flow histogram. The optimal matching pairs of detection and trajectory can be obtained. Finally, trajectory and optical flow histogram model are updated by estimated position and bounding box. Experiments verify the effectiveness of the proposed algorithm. Moreover, we show that our method can avoid false positive, false negative and identity switch problems relative to Kalman filtering method.