Optical Flow Algorithm Via Analytic Wavelet

In this paper, we discuss the detection of moving targets, referring to how to detect moving objects in continuous images. This problem has a strong prospect of practical application in modern life, and is widely used in all parts of areas.Image sequences contain not only the target in a real three-dimensional space under the two projection image plane, but also include the movement trends in temporal dimension. How to use two or more frames to describe the characteristics of the moving targets, is the core problems to be solved in computer vision. In the first chapter, we introduce the basic situation and status of the moving object detection, and make a brief introduction for every chapter.Then we analyze the processing steps of the two and three differential to detect moving objects, and make some related experiments. We find the algorithms have a fast speed in computation, while the target hole and the disturbance of the light intensity will bring analytical obstacles. Subsequently, we make a study of the background difference methods, we give the Gaussian background update and the Gaussian mixture model update, which can greatly improve the lack of inter-frame differences.In this paper we focus on the optical flow model and features. Combined with multi-resolution of the wavelet, we analysis the advantages of the analytic wavelet, it will not bring the cumulative error, and also will avoid bringing phase oscillation of the real wavelet, thus reduce the calculation error. In paper we select the Gabor wavelet, and analysis the time-frequency image of Gabor function. We select synthetic and true moving image sequences to make experiments, and parameters are manually select. In program algorithm, we choose frequency by rotating it, and in the final choices we use a non-maxima suppression strategy, we make an accuracy of optical flow calculation through a numerical evaluation, and get better conclusions.Finally, we make an overall analysis and outlook for these methods. Although the computational complexity is high, the image textures are accurate through our method, which confirms that the Gabor function had a better detection result for the linear edges.