Research On Video Surveillance Image Stitching Of Urban Comprehensive Law Enforcement

With the development of machine vision and the intelligent trend of people's lives,image stitching technology attracts more and more attention.The study of image stitching technology in urban comprehensive enforcement video system provides a more compact and reliable data for other parts of the system.The main steps of image stitching and the overall flow of video panorama stitching are studied as follows:(1)Image registration based on geometry: The SIFT algorithm has the problem of large amount of calculation when calculating characteristic vector and feature point matching.Combined with the advantages of small area matching and the idea of reducing the number of feature points,the corresponding improvement is carried out based on the SIFT algorithm in this thesis.Firstly,we locate the overlap area of images roughly by the phase correlation method,and then extract the feature points of interest in the overlap area.We structure the Voronoi diagram using the extracted points and divide the image into 4 rows and 4 columns.According to the Voronoi,we find four pairs of matching points respectively in every small area and calculate the transformation matrix of the corresponding image,then calculate the image transformation relations with eight transform matrix.Searching for a quantitative point within a specific area reduces the number of the points you want to pair,which increases the efficiency of image stitching.(2)Image fusion based on optimal stitching line: The ghost appears after using weighted fusion based on the overlap area,and large amount of calculation is caused by the fusion based on multiband.The most ideal way for fusion processing is based on optimal stitching line.The local Optima is likely to happen when selecting the stitching line,and inconsistency is caused when cutting objects by stitching line.To solve the problem above,we calculate the color invariants of each pixel in the overlap area of two images to be stitched together.Then,extract the edge of the object in the area and calculate the nearest distance of each pixel.Take the pixels within the region as the vertices of the graph,and combine color variables and the nearest distance as edge weights.The graph composed of the overlap area uses the maximum flow / minimum cut theory to obtain the optimal stitching line.Finally correct the stitching line based on color invariant.Image fusion based on this method has a promotion in the image fusion evaluation indicators.(3)Video panorama stitching: Video has the problem of fuzzy image frame,information redundancy and differences of color and exposure between video frames and so on.The problems can be solved through the introduction of panoramic stitching process one by one.Firstly,the fuzzy frame is removed by the change of the edge pixels in the image.Secondly,the frame difference clustering algorithm is used to find the key frame to get the panoramic image set.Then color correction of the image is carried out.And then use the image registration method above to register the image and unify all the images into the same coordinate system.Exposure compensation is made for registered images.At last,the proposed fusion method is used to get the panoramic image.This treatment makes the quality indicators of panoramic image higher.