Light Weight Attention Module Based Video-Based Person Re-Identification for Real-Time Environment

Person re-identification (Re-ID) is the process of identifying a person among various non-overlapping camera views, where each view may have different illumination conditions, camera angles, occlusions, and resolutions. In a real-time environment, person Re- Idrefers to the ability to perform this identification process in real-time, typically using video streams from surveillance cameras. It involves capturing video from multiple cameras, extracting features from each video stream, and then matching the features across multiple streams to identify a specific person. It requires fast and accurate algorithms that can handle large amounts of data in real-time. These algorithms typically use deep learning techniques, such as convolutional neural networks (CNNs), to extract features from the images or videos of individuals, which are then used to match individuals across different cameras. The performance of person Re-ID algorithms rely on the quality of the input data, the intricacy of the scene, and the accuracy of the matching algorithm used and efficiency or speed. In this paper a lightweight attention module has been proposed to deal with re-identification process in real-time environment and achieved a remarkable mAP of 90%, rank-1 accuracy of 92.7% with a speed of 0.0399 sec/batch.