The Research Of Variational Optical Flow Model Based On Fractional Order And Its Application On Video Monitoring System Of Escalator

Motion detection is one of the fundamental task in image processing field and the cornerstone of complicated image processing projects.The optical flow algorithm is the most popular motion detection methods.that obtain the speed and direction of every pixel in an image.Variational optical flow method has been used widely due to its several featuers that are complete mathematical theory,clear structure,powerful compatibility and so on.However,basic assumptions for the variational optical flow method are often violated in practical applications such as illumination anomaly,severe occlusion,noise interference,large displacement motion,multi-motion and non-rigid motion.These violations result in a large deviation between the estimated motion and the actual motion.To cope the various practical problems(illumination changes,low contrast and uneven brightness),a lot of improvements have been made to the variational optical flow model based on the fractional order theory.Focusing on the illumination changes problem in optical flow estimation,a dual fractional order variational optical flow model(DFOVOFM)is proposed.In particular,the fractional-order derivative is applied in the brightness and smoothing constraint equation of HS model.The Euler-Lagrange equation is utilized to the minimization of the energy function of the fractional-order optical flow model.Two dimensional fractional differential masks are proposed and applied to the calculation of the model simplification.Compared to conventional variational optical flow model,our model is more robust to environmental changes.Furthermore,the optimized calculation method makes the model easy to be realized by computer programming.Aiming at accurate optical flow estimation in low contrast region,an adaptive dual fractional order variational optical flow model(ADFOVOFM)based on DFOVOFM is proposed.The main innovation of this work is to fit a flow field regional to a variety of fractional order differential masks and the domain of each region is determined adaptively.The order and size of the fractional order differential masks for each region are adjusted by image signal to noise ratio(SNR)while the shape of the fractional order differential mask is regulated to prevent interference from surrounding regions.Adjusting the fractional order differential mask adaptively enables the proposed method to accurately segment motion objects in low contrast region as well.Targeting on accurate optical flow estimation in scenes with illumination or noise asymmetry,an adaptive smoothness parameter strategy(ASPS)based on image quality parameter is proposed.First,an algorithm combining simple linear iterative cluster(SLIC)with local membership function is used to segment the entire image into several super-pixel regions.Then,the image quality parameters of each super-pixel region consisting of colourfulness,contrast and signal to noise ratio are calculated respectively.Finally,a multiple perception model is applied to calculate the smoothness parameter by the image quality parameters of each super-pixel region.Allocating different smoothing parameters to the superpixel region with different image quality parameters enables the proposed method to accurately estimate optical flow fiend in scenes with illumination and noise inhomogeneous as well.The abnormal behaviors detection and recognition of pedestrian have always been a challenging task in intelligent video surveillance system.To cope this problem,an algorithm combining ADFOVOFM+ASPS with the OpenPose based skeleton extraction method is proposed.At first,ADFOVOFM+ASPS is used to estimate the optical flow field under scenes with illumination changes,low contrast and uneven illumination.At the same time,the OpenPose based skeleton extraction method is used to detect the location of passengers on escalator.Then,the optical flow field and the human skeleton are combined to obtain the velocity and direction of the passenger head.After that,the optical flow field of the passenger head and the escalator step under the passenger foot are used for abnormal behavior detection and recognition.Experimental results show that the proposed model and its improvement strategy can accurately estimate the optical flow field in real time of low contrast outdoor videos with insufficient illumination,uneven brightness and illumination changes,the accuracy of abnormal action detection and recognition can reach to 97% and 92%.Hence it concludes that our work have significant contribution in the research and application of variational optical flow model.