Research On Moving Objects Detection And Tracking In Methods Intelligent Visual Surveillance System

At present, intelligent video surveillance is facing many problems in the research field and application, but it also make significant achievements. Now with the support of state policies and the development of computer vision technology, moving target detection and tracking is playing an in-depth and extensive role in many fields. Moving target detection and moving target tracking are the two most core and critical steps for intelligent video surveillance, many institutions and scholars propose and implement related algorithms for different application areas, but face with the complex and changing environments, there still many problems to be solved in moving target detection and moving target tracking. In this paper, aiming at the existing problem of key part of video surveillance, detection and tracking of moving target, we put forward the corresponding solutions. The main work is as follow:In this paper, we propose the corresponding improvement methods to LBP texture operators, which is commonly used in computer vision technology. And for a large number of local texture feature is a fraction of video types described by the LBP operator characteristics. Therefore, we merge a large number of invalid LBP operators as a class, and design a new LBP texture feature descriptor, so that the operator species that description of local texture feature graphics is reduced to58kinds. This paper presents all the circumstances presented by the58species operators. And combining with practical application presents8-bit binary LBP coding to LBP unified pattern conversion code table. So that the time complexity of the shadow suppression and characteristics matching algorithm for moving target detection and tracking significantly reduce, and increase the speed of operation of the algorithm.For the difficulty of widespread shadow suppression in moving target detection algorithm, this paper proposes corresponding solutions based on Gaussian mixture background model. Through a detailed study and analysis of the video background information, after processing the foreground information which is processed by the Gaussian mixture model, with the LBP texture and color information, we find that this method can better suppress shadow after experiments. The basic idea of this algorithm is, at first, using Gaussian mixture model to detect the moving foreground, then we combine the cache to calculate the foreground part of the LBP texture feature, watching them matching or not. And combining RGB color information to detect the real foreground. Removing shading effect on target detection algorithm, experiment results show that the improved algorithm of moving target detection has obvious effect on shadow suppression.This paper takes deep and careful researches into currently hot topic of kernel density tracking algorithm, and proposes improved method to single target feature detection algorithm of kernel density tracking algorithm. Paper presents a multi-feature kernel density tracking framework, the framework can adapt to multi-feature characteristic description of moving target. The concrete idea is to establish the specific characteristics of sub-model of kernel density tracking. Using the linear superposition of target sub-model created by each characteristics description to express the target model and candidate model of kernel densities. Combined with the log-likelihood function to establish a way to update each feature sub-model’s dynamic weights. Implementing the priority of better discrimination of background characteristics with the tracking of multi-feature kernel density. Making up the disadvantage of not fully enough to the single moving target characteristic describes. Paper combines two commonly used features of LBP texture an RGB color in video analysis, and verifies multi-feature kernel density tracking algorithm, experiments results show that tracking effect is good.