A Study On Brightness Consistency-based Fast Image Fusion Technology

Video surveillance systems have developed rapidly in recent years and the amount of data has increased tremendously.These huge amounts of data not only bring challenges to human retrieval and browsing,but also bring tremendous pressure on data transmission and storage.Therefore,intelligent video surveillance technology has become a research hotspot.It adopts more flexible compression processing and reconstruction methods to achieve lower network bandwidth consumption,and condensed redundant video with a shorter video summary.As one of the key technologies,image fusion will affect the visual effect after fusion and to some extent affect the performance quality of the final video.Technically,fusion can be divided into two parts:target extraction,foreground and background synthesis.At present,the accuracy of common target extraction algorithms based on video sequences is not enough,which does not meet the requirements of fusion algorithms for foreground integrity.Based on this,this thesis studies and designs a set of fast image fusion algorithms to meet the needs of complete targets,low overhead,and seamless fusion.The main work includes the following two aspects:(1)An improved ViBe moving object extraction algorithm based on saliency analysis is studied.ViBe(Visual Background Extractor)algorithm is widely used with its advantages such as simplicity and speed.However,in the detection process,due to the limitation of the number of samples and the random selection of samples,fragments and holes are prone to appear.The loss of the foreground edge does not meet the accuracy requirements of the fusion algorithm.Therefore,based on the fusion,this thesis makes improvements to the ViBe algorithm for this phenomenon..Debris generated during extraction was removed by morphological processing.Aiming at the problem of large-area holes in the target,a method for filling holes based on visual saliency analysis was proposed.First determine the range of the area to be detected,and use the significant value to calculate the similarity function,analyze the attributes of the pixel block to be detected,and obtain an accurate foreground target.Through experimental tests,it is proved that the improved algorithm improves the accuracy and recall by about 2%-10%compared with ViBe.(2)The optimization of a hybrid fusion technology based on brightness consistency analysis is studied.The weighted fusion algorithm is efficient,but it is only a simple superposition,and gaps are prone to appear at the fusion boundary.The Poisson fusion algorithm works well,but when solving the linear Poisson equation,the migration of the gradient field will cause the bleed phenomenon,and the complicated solution process will bring too high time complexity.On this basis,this thesis uses the color space conversion to improve the bleeding of Poisson fusion,and analyzes that the biggest difference between the backgrounds before is the difference in brightness,making full use of the advantages of the weighted fusion algorithm and Poisson fusion algorithm,and using matrix singularity The value calculates the brightness scale factor,adaptively adjusts the brightness difference between the backgrounds and selects an appropriate algorithm to improve the efficiency of the algorithm while meeting the fusion requirements.The experimental test proves that the improved fusion algorithm basically meets the seamless requirements.Compared with Poisson fusion,the efficiency of the fusion algorithm is improved by 1.5-2 times.Finally,this thesis uses the CDnet2014 dataset to test the algorithm.Finally,experimental results show that the fast image fusion algorithm based on the background model proposed in this the thesis basically meets the requirements of real-time,accuracy,and low computational cost of the monitoring system on the processing results.This thesis has explored and tried foreground extraction and adaptive,fast and efficient image fusion methods in video surveillance scenes,which has certain value for the improvement and improvement of better image processing.