Research Target Tracking Based On Multi-fisheye Lens Image

Target tracking has always been a hot research direction in the field of computer vision.Many researches on target tracking algorithms are based on a single camera.In practical applications,the visual field limitation of a single camera cannot meet people’s needs,so target tracking under multiple cameras has aroused people’s interest.Multi-camera target tracking has important research value in intelligent monitoring,military and virtual reality,etc.Multi-camera collaboration also promotes the development of some cutting-edge technologies in computer vision,which brings a lot of convenience to human life.It is worth the great efforts of scientific researchers.This paper mainly studies the target tracking based on multi-fisheyes lens.Firstly,this paper analyzes the imaging principle and distortion model of fisheye lens,and improves the corner detection method based on the checkerboard calibration algorithm under limited experimental conditions to calibrate fisheye lens.Secondly,the paper analyzes the three kinds of image feature point detection technology,and experiments respectively different feature point detection algorithm of detection results.On the basis of the feature point detection,in order to improve the accuracy of feature points matching,this paper uses RANSAC algorithm,chunking overlap,purified signs point,reduce the number of iterations matching,calculation homography,and after purification experiment validation feature points matching precision is higher.In this paper,in view of the ghost effect caused by moving objects in overlapping areas during mosaicing,a texture difference factor is introduced into the optimal suture line estimation,and a new suture line estimation energy function is defined.On this basis,the improved fusion algorithm is improved,and the effectiveness of the improved fusion algorithm is verified by fish-eye image.Then the traditional nuclear related filter target tracking algorithm theory and mathematical foundation,and analyze its actual tracking target dimension and keep out the problems,based on this to improve traditional nuclear related filtering algorithm,this paper for the target scale problem,the paper calculated the target values of maximum response to determine whether a target dimension change and adjust the size of the target frame,to the problem of target block based on the target response value information to determine the coordinates of corresponding object shelter,and filter design window predictor to predict the location of the object.And compare the effect of target tracking before and after improvement through experiment.Finally,based on the position of multi-fisheye lens in this paper,the multi-graph stitching strategy is proposed,and a target tracking scheme based on multi-fisheye lens is designed.The experiment of this scheme is carried out,and the target tracking under multi-fisheye lens is completed.The tracking scheme in this paper uses the fish-eye lens with a wide field of view to reduce the number of cameras and obtain a larger field of vision.In addition,the tracking scheme avoids the problems of target consistency identification and target handover,and reduces the complexity of target tracking.