基于支持向量机增量学习和LPBoost的人体目标再识别算法

Abstract

Object association between the cameras is a key of persistent object tracking in non-overlapping multi-cameras.A people re-identification algorithm was proposed only using people appearance completely independent on the space-time relations,object association across disjoint views was carried out by directly utilizing identification result,and this method does not depend on captured time of object and path restrictions.Complementary visual word tree histogram and global color histogram were extracted from the video image sequence,and Support Vector Machine(SVM) incremental learning was used to train online distinguishing people appearance models of two features.Finally,multi-class Linear Programming Boosting(LPBoost) algorithm was introduced into on-line adaptive fusion of two SVM models.The proposed method has strong online learning ability,and can incrementally represent discriminative people appearance model.The model after fusing two features is more discriminative and effectively reduces the influence of changes in various conditions.Experimental results show that the proposed method achieves high identification rate and rapid real-time implementation which are markedly improved compared to the existent methods.