ABSTRACT
People re-identification is an essential piece of intelligent video surveillance systems; also, it is crucial for several more complex tasks such as scene understanding, person tracking, and activity recognition, among others. Nowadays, significant advances are reported towards solving the task. Nevertheless, when the methods are tested under real conditions, such as several cameras, several poses and angles, zooms, etc., their performances decrease considerably. Therefore, it is important to evaluate the performance of the methods under the same conditions and real scenarios. The idea behind this, is to show the current state-of-the-art in this particular task, and also, the opportunities for improvements.
This chapter is devoted to measuring, in real environments, where we are and how far we are in solving the people re-identification task, as well as having a clue about what approaches have more potential for improvement in near future. A concise summary of the recent contributions in the person re-identification task is presented here after studying the existing research articles focusing on Cumulative Matching Characteristic procedure and tested on the more used and cited benchmark datasets.
