Smart surveillance system for real-time multi-person multi-camera tracking at the edge

Abstract

In this work, we have presented an end-to-end multi-person multi-camera tracking (MPMCT) surveillance system and implemented it on edge analytics platform for real-time performance. The proposed MPMCT framework is both privacy-aware and scalable supporting a processing pipeline on the edge consisting of person detection, tracking and robust person re-identification. A realistic and large dataset has been created to train and evaluate the surveillance system that has been employed to track people inside the institute campus throughout the entire day. Appropriate deep-learning algorithms and real-time implementation strategies have been employed to realize the MPMCT system on NVIDIA Jetson TX2 embedded platform with real-time performance. The proposed system has an IDF1 score of 90.97 on our dataset and outperforms the current state-of-the-art real-time algorithms. The performance up to 30 FPS is achieved for the person detection algorithm, whereas an average latency of 90 ms is achieved for the re-identification algorithm.