Research On Moving Target Detection Algorithm Based On Road Video

Along with the rapid development of economy and the continuing social progress,the road traffic as the third industry,links more and more close with people's daily life or the national economy.Many countries in the world are carrying out large-scale road construction,at the same time,they also have carried out the research of intelligent traffic management,achieved a lot of scientific research achievements and successful experience,besides,obtained a perfect road monitoring management effect.By the results of various studies and algorithms proposed by many papers,we can understand that,if we want to get further development of intelligent road monitoring,a moving target detection algorithm with real-time performance,high detection precision,good robustness is necessary.The Gaussian Mixture Model algorithm,as a classical and effective detection algorithm of moving target,has been widely studied and applied,however,the algorithm has some defects,as the same as most moving targets detection algorithms.Therefore,this thesis presents the improved algorithm which is used in road traffic situation.At first,this paper analyzes the basic image processing knowledge related with the moving target detection technology,including gray image,image filter processing,mathematical morphology processing,image binarization,meanwhile,introduces some typical methods in the moving target detection fields.Also,this thesis introduces three kinds of typical moving target detection algorithms including their basic principles,application scenarios,advantages and disadvantages.Focuses on the Gaussian mixture model,this paper introduces its basic principle and analyzes its important parameters,then,two improved Gaussian mixture model target detection algorithm is proposed in view of its disadvantages in the road video moving target detection:(1)We establish basic Gaussian mixture model,and introducing the illumination correction model,then,we combined the two model in order to solve the phenomenon of false detection caused by the sudden change of illumination.on the detection results,we apply a series of treatment(operation of mathematical morphology,connectivity analysis,etc.),to achieve the purpose of optimizing the detection effect.(2)Gaussian mixture model often misses the targets when it is used in road video including the gradually stationary targets.In view of the defects of Gaussian mixture model,we introducing the background learning parameters which can effectively improve the phenomenon of missing targets.This paper also introduces the ZedBoard development platform.We describe the process of building its software environment,and give the experimental results and performance analysis of the improved Gaussian mixture model on the platform.