Development of an Improved Occluded Person Re-Identification System Using Deep Learning.

One of the core challenges faced in developing a person Re-Identification (re-ID) model for open-set is solving the problem of intentional or unintentional occlusion of the subject or person of interest. There has been numerous research conducted on closed-set re-ID where the basic assumption is that the gallery set will only have non-occluded person images. This is not practical for real-world scenarios where the variables vary unpredictably. This paper examines the problem from an object detection perspective as well as surveys the state-of-the-art of person re-ID, their achievement and shortcomings. This paper has also brought into light the recent advancement in object detection algorithms outside person re-ID and how those techniques and ideas can help within the realm of the person of interest search. The aim is to realise a person re-ID system that is geared towards open-set Re-ID and can deal with person occlusion in a robust manner.