PTZ Camera Teams for Effective, Close In Multi Object Detection and Tracking

In this study, we introduce the concept of a camera team for real-time, multi-object detection and tracking using pan-tilt-zoom (PTZ) cameras. Traditionally, the use of PTZ cameras for the detection and tracking of multiple objects has been constrained by the zoom-field of view (FOV) trade-off. Detecting and tracking two distant objects typically necessitated a broader FOV, resulting in a reduction in zoom capabilities. The proposed camera team comprises multiple PTZ cameras with nearly coincident optical centers, where each camera is dedicated to tracking a single target exclusively. We employ YOLOv7 and DeepSORT algorithms on individual cameras for target detection and tracking, respectively. Additionally, an auxiliary PTZ camera can be utilized to provide an overarching view of all targets, if required. The central control code orchestrates all camera movements in a synchronized fashion, facilitating various operational modes including ‘chase’, ‘solo track’, and ‘correlate’. Experimental results demonstrate that this camera team architecture significantly enhances the efficiency of object detection and tracking across expansive areas, providing high-resolution feeds of multiple targets concurrently—a feat unattainable with traditional camera setups.