A Dual-Camera-Based Ultrafast Tracking System for Simultaneous Multi-target Zooming

Abstract

In this paper, we develop a novel dual-camera system that can simultaneously capture zooming-in targets by combining an ultrafast pan-tilt camera and a fixed wide-view camera. According to the positions of all the targets recognized with deep learning in the wide-view camera, the pan and tilt angles of multiple pan-tilt cameras are controlled virtually on the ultra-fast pan-tilt camera through multi-thread viewpoint control to simultaneously capture the zooming-in images of all the targets. Our system can generate five hundred virtual cameras with different zooming views in a second, and the effectiveness of our system is demonstrated by showing experimental results in simultaneous zoom shooting for multiple running persons and cars in the range of 70 m or more in a natural outdoor scene.