Person Re-Identification

Person re-identification task consists of recognising images of a person of interest across several non-overlapping cameras. It is a challenging task since it presents several issues such as different poses, several illumination variations and camera resolutions, different backgrounds, and several occlusions.

This task can be helpful in reconstructing the movement of a suspect individual. The main goal of a person re-identification system is to reduce the time required for a human operator to analyse all the available videos. 

The recorded videos (acquired by network cameras) are shown to the user. Simultaneously, these videos are processed by the pedestrian detection and bounding box extraction module which aims at detecting and extracting pedestrians. The extracted images are, then, stored in a database, called gallery. 

The user can select (from a video) the person of interest (query) through a user interface. Once the query is selected, the features of query and gallery images are extracted by the feature extraction module, and compared by the similarity measure module. 

The output of the system is shown to the user. It consists of a list of images ordered by the similarity to the query.