A networked high-speed vision system for vehicle tracking

Abstract

This paper presents a networked vision system for tracking high-speed objects moving across multiple cameras' fields of view. Tracking vehicles that travel along highway is one of the potential applications, and the vehicles can be continuously tracked from the entrance to the exit without being lost. Such a system can be used for surveillance and analysis of traffic congestions/accidents. Each vehicle passes a large number of cameras at high speed, one ofter another. Higher frame rate and faster data communication are desired for reliable target tracking and system scalability. We developed a prototype system that captures the moving vehicle with 1000 frames-per-second and that shares the small feature data only between a pair of adjacent cameras. A 1/10-scale vehicle moving across two cameras at 2500 mm/s ≈ 9 km/h was successfully tracked by the experimental system. Tracking a laser pointer spot moving at faster than 100 km/h on the floor is also demonstrated.