Automated Person Search in a video using Face based Person Re-identification system

Intelligent surveillance is an important computer vision task, that can be used in real-time to monitor public places like airport, temples, mall etc. It is essential, since person of interest involved in crime can be found and tracked using surveillance cameras. Manual monitoring of surveillance videos is time consuming and boring to watch 24x7. Hence computers should be automated to ease this task. Same person captured across multiple cameras are associated using features that are common. The process of enabling correspondence between similar images that are taken in different lighting, different background, different costume, different pose and different time is known as person re-identification. Various methods have been proposed by researchers to find a solution to person re-identification problem. The main objective is to generate robust discriminant features from the entire human body images of the same person with varying pose, background and illumination. The results are further refined using facial key points of the face detected from the matched images. Deep Learning layers are customised to meet the objective. The experiments are carried out using CUHK03, a benchmark dataset. The model is tested using a real-time data collected at SRM Institute of Technology. The results are compromising and screenshots are included.