A New Person Re-Identification method by Defining CNN-Based Feature Extractor and Sparse Representation

Abstract

Person re-identification (re-id) is one of the most important and challenging topics in image processing and artificial intelligence. In general, person re-identification means that a person seen in the field of view of one camera can be found and tracked by other non-overlapped cameras. Low resolution frames, high occlusion in crowded scène and few samples for training supervised models lead re-id to be challenging tasks. In this paper, a new model for person re-identification is proposed to overcome the noisy frames and to extract robust features from each frame. To this end, a noise-aware system is implemented by training auto-encoder on artificial damaged frames to overcome on noise and occlusion, and model for person re-identification is implemented based on deep convolutional neutral networks. To evaluate the proposed method against general methods Rank-k is used for different k’s. Experimental results on two important databases, CUHK01 and CUHK03 demonstrate that the proposed method performs better than state-of-the-art methods.