Moving Object Detecting And Tracking Technology Based On Improved Optical Flow Method

Moving object detection is an important part in the research of machine visual, by acquiring the parameters of the moving target such as position, velocity, and acceleration, the target trajectory can be predicted and analyzed, then the moving target can be classed, recognized and tracked. The detection and tracking algorithm is influenced by the external factors such as shape and background change, environmental noise, and occlusion, so it is widely accepted that the measurement of the algorithm is through real-time, robustness and accuracy. The detection and tracking technology based on the optical flow estimation algorithm has high accuracy, and the moving parameters can be obtained completely and directly, but its real-time performance can not meet the practical application, it needs to be improved.In this paper, the main target detection algorithms are discussed:background subtraction algorithm, inter-frame difference and optical flow algorithm, the advantages and shortcomings of these algorithms are compared. According to the optical flow and inter-frame difference algorithm, an improved optical flow estimation algorithm based on the frame difference is proposed.First, in the part of frame difference, the discontinuous frame difference is applied to detect the moving target with the inter-frame displacement smaller than one pixel and accumulated displacement bigger than one pixel. Then in the part of calculating the optical flow, the general dynamic image model (GDIM) is introduced to form a new optical constraint equation, which has overcome the problem that the equation is not tenable. Only not zero pixels are calculated by using the optical flow method. As a result, accuracy and speed of the target detection are improved. The effectiveness of the proposed method is verified by the simulation results.In the part of target tracking, the improved optical flow model and kalman filter algorithm is used to track the moving object. Through the kalman filter, the position of the target in the next frame can be forecasted, the velocity and direction obtained from the optical frame detection can effectively reduce the searching range, which can realize the fast tracking of the moving target.