Research And Implemention Of Search And Track Algorithm Based On People Attributes

Detecting and recognizing people is a key field of computer vision technology. In thepractice of surveillance, face recognition, which has been proved to be effective to identify aperson, can only work under some strictly controlled scenarios. Besides, there is always no moreinformation about the target person except some attribute descriptions from witness underemergency scenarios. But rather than getting images at the face level, some attributes are alwaysobservable at a distance, such as gender, hair, clothes, ect. If a surveillance system couldcombine all the attributes to searching the person of interest, it will be a significant approch inthe application of security.This paper proposes a novel system for searching person of interest based on peopleattributes, and an algorithm implementation based on the framework. This algorithm detectssome attributes of a person, such as gender, hair and glasses, extracts and models them as anattribute vector, then search the videos to find the target person with a specified attribute vector.Experiments has been presented to evaluate the performance of the proposed system on thevideos from real surveillance environment, the results demonstrated that this system has mainlyachieved the expected goals of locating and tracking a suspect in multi-camera surveillance.