Parallel Optical Flow Calculation And Object Detection Based On Markov Random Field

Optical flow refers to the instantaneous displacement velocity of a pixel on a two-dimensional plane projected from a three-dimensional moving object.It not only contains the moving direction and displacement information of the moving object,but also carries the three-dimensional structural information of the target.Therefore,the optical flow is often used to study motion of project.Optical flow is a hot research topic in the field of computer vision and image processing.With the rapid development of computer technology,optical flow computing has been widely used in such fields as military,aerospace,transportation and medicine.In recent years,with the deepening of optical flow technology research,various optimization strategies and methods have been applied to the optical flow calculation model.Although the calculation accuracy of optical flow has been improved steadily,the computational efficiency is reduced due to the higher computational complexity of optical flow.In this paper,a variable optical flow calculation model based on Markov Random Field(MRF)is proposed and implemented by parallel belief propagation(BP)algorithm,which greatly improves the computational efficiency.Finally,the parallel optical flow method and the parallel level set algorithm are combined to design a target detection model based on MRF optical flow,which realizes fast detection.The main contributions of this article are:1.Aiming at solving the time-consuming and robust problem of optical flow calculation,a parallel optical flow calculation model based on MRF is proposed.In the process of optical flow calculation,the classical optical flow model is transformed into an optical flow model based on MRF.The gradient conservation and non-square penalty function strategies are added to the classical hypothesis to overcome the problem of illumination changes in the optical flow field.It can filter out noise and optical flow anomalies well.Therefore,the accuracy and robustness of the optical flow calculation are improved.In order to speed up the calculation,the belief propagation technique is used to minimize the energy function of the MRF optical flow,and the optimized belief propagation is designed according to the characteristics of the belief propagation algorithm on the CUDA platform.Finally,using the data set provided by Middlebury database as the experimental image,the proposed algorithm is compared with the classic Horn algorithm,Weickert optical flow algorithm,Hossen's parallel Lucas algorithm and parallel optical flow algorithm based on Markov random field proposed by Grauer-Gray et al.The experimental results show that the proposed algorithm is robust to scenes of illumination,noise and motion discontinuity,and the computational speed is greatly improved.2.In order to achieve fast target detection,the level set method is combined with the optical flow model proposed in this paper to realize the target detection methodbased on MRF optical flow.Firstly,the initial closed contour curve is obtained by optical flow calculation.By calculating the distance from the pixel point of the image to the curve,the shortest directional distance(ie,the level set function)of each pixel point in the image to the initial curve is obtained.Then,the directional distance of each pixel is continuously updated according to information such as image gradient,curvature,etc,by extracting pixel points with a directional distance of zero,and finally obtaining a closed curve close to the edge of the target contour.In order to solve the computational time-consuming problem of reinitializing the level set function in the level set algorithm,the GPU's CUDA acceleration technology is used to further improve the computational efficiency.The algorithm is verified by the real traffic road image sequence.The experimental results show that the fast target detection can be completed well.