Novel Attention-Based Framework for Person Re-identification in Video Surveillance

Intelligent video surveillance systems, automatic entry and retail systems at theme parks, airport passenger flow control, and automated driving behavior analysis are a few applications where critical insights can be generated by detecting and identifying people through a camera network. In different public locations, such as train stations, airports, hospitals, shopping centers, etc., widespread camera networks are used daily, covering vast areas and having non-overlapping perspectives and providing a tremendous amount of relevant data. We cannot rely on manual monitoring to use this data efficiently for public safety applications and thus need reliable automated systems that can track the behavior of a person through several cameras. Person reidentification (PReID) is a simple task for this and plays a vital role in re-identifying persons from video surveillance cameras. In this paper, we leverage the attention-based mechanism to re-identify persons from video surveillance cameras by using the image modality translation and CycleGAN, which served as data augmentation in our proposed network. We outperformed against all measures compared to other state-of-the-art methods.