Person Re-Identification Under Intelligent Video Surveillance

With the development and application of video surveillance,its role in today's social life is also increasingly important.In the field of computer vision,the processing of pedestrians in video screens has always been the focus of researchers in various countries.The development of person recognition is also deroved from the idea of search for person in surveillance video.The main task of person re-identification is to judge whether a target person has a matching person in non-overlapping camera view.Due to a series of problems such as changes in lighting,pedestrian attitude,background differences,and camera parameters during the process of matching different persons,research on person re-identification has always been a challenging research topic.The specific applications of person re-identification is starts with the three directions of person detection,feature extraction,and metric learning.The main work of this thesis is as follows:1.In the pedestrian detection part of the video,this thesis first designed a foreground extraction scheme based on Gaussian mixture model to detect the human body in the foreground region with a moving target.And the HOG feature was used in combination with the classifier to detect the human body.At the same time,we compared another pedestrian detection based on the DPM model,although the human detection by the DPM algorithm is better than the former,it can not meet the application in real-time processing.Therefore,this thesis mainly adopts the human detection method of HOG features combined with foreground detection as the detection work of this paper.2.In the person re-identification,this thesis first use a Local Maximal Occurrence Representation to obtain robust person features.This feature extraction method can firstly satisfy the person re-identification system at the feature extraction speed.And then this feature extraction can also be robust to the linghting changes and the change of camera view.At the same time,this paper also adopts the saliency feature to improving the accuracy of person re-identification?3.Finally,in the design of person re-identification application system,this thesis integrates the above-mentioned pedestrian detection and pedestrian re-identification modules to realize a person target that can facilitate to the user to quickly obtain the same person from the surveillance video.