ByteReID: An Efficient Online Multi Camera Multi-Person Tracking System

• Main Authors: Bergia, Stefano, Rizzo, Giuseppe, Gaetani, Manuel, Chiesa, Mario, Dominici, Fabrizio
• Format: Conference Proceeding
• Language: English
• Subjects: Accuracy; Cameras; Faces; Identification of persons; Information and communication technology; Lighting; On-line feature embedding clustering; Online multi-camera multi-object tracking; Overlapping cameras; Real-time systems; Robustness; Video surveillance
• ISSN: 2472-8586
• DOI: 10.1109/AICT61888.2024.10740451

Current approaches for person re-identification suffer from long computing time despite having good accuracy scores and this translates into low applicability in real world environments. In this work, we introduce an efficient online multi-camera multi-person tracking system designed to speed up surveillance capabilities through a two-step hierarchical approach. This approach, called ByteReID, integrates efficient tracking within individual cameras and employs an advanced online clustering mechanism for effective association of tracklets across multiple cameras. Central to its design is a specialized feature encoder that enhances the system's ability to distinguish between individuals, promoting accurate tracking without initial knowledge of the number of subjects nor their features. It is designed to be flexible across various surveillance setups, ensuring robustness against challenges such as differing camera orientations, lighting conditions, and overlapping fields of view. ByteReID is optimized to operate in near real-time setting and its methodology offers an efficient near real-time solution to the demands of video surveillance, ensuring wide applicability across different operational environments.