Research On Car Washing Behavior Recognition Methodbased On Multi-core Multi-instance Learning

Washing behavior recognition is a branch of human behavior recognition under complex scene,now,the simple recognize of simple human act scenario has basically been resolved,and behavior recognition complex scenarios still faces many difficulties.Special washing line camera angle shooting is not clear outline of human action and automobiles,workers frequently move caused "Ghosting Region" makes identification more difficult for car washing behavior.The current conventional behavior recognition algorithms and can not adapt to this special environment washing line identification for car washing behavior,learning algorithm is proposed based on a multi-core multi-instance,in order to improve the accuracy of the workers at the car wash car wash line environment behavior recognition.In this thesis,an improved ViBe background subtraction detection of moving objects in real time to solve eliminate "Ghosting Region" problem,use HOG-LBP feature extraction algorithms to deal with human actions outline the problem is not clear.The identification algorithm uses multi-core multi-instance learning algorithm that would multi-core support vector machine learning algorithm and the combination of multi-instance,can effectively deal with the integration of feature extraction to HOG-LBP,but also improve the learning ability of recognition algorithms to further improve the car wash behavior recognition accuracy.The experiments show that,multi-core multi-instance learning algorithms on experimental data set to build car wash behavior,compared with the traditional behavior recognition algorithms,its recognition rate increased.In this paper,an algorithm is proposed for the washing line environment,also applies to the behavior of a complex scene recognition problems similar environment under the washing line.