Embedded Vision System for Real-Time Object Tracking using an Asynchronous Transient Vision Sensor

TLDR

The CMOS temporal contrast vision sensor asynchronously responds to illumination changes with a 120 dB dynamic range and sub‑100 µs latency. The paper introduces an embedded vision system and a 1 ms‑resolution tracking algorithm for real‑time object tracking using the 128×128 pixel sensor. The system encodes events as address‑event representation data, processes them on the imager’s focal‑plane chip, and runs the tracking algorithm on a low‑cost, low‑power DSP. Real‑time vehicle tracking in traffic monitoring was demonstrated, and the method also shows promise for people tracking.

Abstract

This paper presents an embedded vision system for object tracking applications based on a 128times128 pixel CMOS temporal contrast vision sensor. This imager asynchronously responds to relative illumination intensity changes in the visual scene, exhibiting a usable dynamic range of 120 dB and a latency of under 100 mus. The information is encoded in the form of address-event representation (AER) data. An algorithm for object tracking with 1 millisecond timestamp resolution of the AER data stream is presented. As a real-world application example, vehicle tracking for a traffic-monitoring is demonstrated in real time. The potential of the proposed algorithm for people tracking is also shown. Due to the efficient data pre-processing in the imager chip focal plane, the embedded vision system can be implemented using a low-cost, low-power digital signal processor