An Efficient Person Re-Identification Method Based on Deep Transfer Learning Techniques

Abstract

Person re-identification (re-id) is a significant process in applications of video analysis. Several applications in different areas such as airports and stations are used multiple cameras in different places for monitoring and investigation, which are expensive and can be easily abused. Therefore, automatic person re-identification techniques are highly required. The main issue of this field is to find distinguishing features that represent the person. In this paper, we proposed an efficient method to extract the main features based on the deep transfer learning technique for a person re-id system. In addition, we employed a support vector classifier (SVC) as a separated classifier for the final decision to increase the accuracy of the system. We employed several publicly available datasets, which are the main datasets used for person re-id purposes in the literature. The proposed method achieved the best accuracy of 89.59% for rank-1, which outperforms the state-of the-art methods. Finally, the simulation results reveal that the proposed system is efficient prior to person re-id. Keywords— person re-identification; transfer learning; SVC; deep learning; video analysis.