Research On Key Technology Of Abandoned Objects Detection Invideo Surveillance

The original method of using manpower to monitor can not already keep upwith the demand，in order to achieve more accurate and efficient technology ofvideo surveillance，intelligent video surveillance is booming．The technology ofdetecting abandoned objects is an important branch of intelligent videosurveillance， it is a cross technology integrating image processing， patternrecognition，computer vision technology and other subjects．Detection of abandonedobjects uses technology of combining surveillance equipment and image processingtechnology，which makes computers and embedded hardware have the ability ofanalyzing video， thus it can detect abandoned objects automatically andalarm．Detection of abandoned objects has practical significance in everyday life，it can be used to exclude a dangerous situation or to help the owner find lostobjects．Therefore，research on key technology of abandoned objects detection invideo surveillance has important practical significance．In this thesis，the key technologies used in the process of detecting abandonedobjects is researched． It is mainly divided into four parts including basicknowledge，target detection，target tracking and human identification．Basicknowledge includes the color image space conversion，image filtering，shadowremoval and threshold segment technology of image and so on．Methods of imagefiltering include median filtering，mean filtering，Gaussian filtering，morphologicalfiltering and so on，which are contrasted in this thesis. Methods of removingshadow including the method based on color features and the method based on edgedetection are researched and the relevant experiments are conducted．These worklays foundation for subsequent studies．Three methods of detecting target including light-flow method， framedifference method and background subtraction method are contrasted in thisthesis．It is founded that background modeling methods have advantages ofextracting more complete target．Background modeling methods including medianbackground modeling method，the mean background modeling method，singleGaussian background modeling method and Gaussian mixture background modeling method are researched and contrasted．The median background modelingmethod and the mean background modeling method can be achieved easily and hasfaster computing speed，but they are only applicable for easy scenes．Gaussianmixture background modeling method has stronger anti-interference than the singleGaussian background modeling method．When traditional Meanshift algorithm is used for tracking target error is large,the Kalman filter algorithm can effectively estimate the position of the target in thenext moment，so Kalman filter algorithm is used to improve traditional Meanshiftalgorithm，the improved algorithm solve the problem of poor tracking result usingtraditional Meanshift algorithm when color feature of targets are not obvious,movement speed of targets is faster and occlusion occurs．The possible interferencefor the abandoned objects detection caused by human is considered in this thesis，amethod based on merging connected regions is adopted to identify human，whichworks by judging connected regions and depth-width ratio of connected regions，this method is simple and fast．...