Neuromorphic Real-Time Objects Tracking Using Address Event Representation and Silicon Retina

Description

This paper presents a hierarchical neuromorphic system for tracking objects. We use AER (Address Event Representation) for transmitting and processing visual information provided by an asynchronous temporal contrast silicon retina. Two AER processing layers work in cascade for firstly detecting different objects, and secondly tracking them even with crossing trajectories. The output of the system offers not only the position of the tracked object but also the speed in pixels per second. The system is fully hardware implemented on FPGA (Spartan II 200), which is part of the USB-AER platform developed in part by authors. A 97.2% of the Spartan II is used for 128x128 pixels input resolution and 6 maximum objects recognition and tracking.

Abstract

Ministerio de Ciencia e Innovación TEC2009-10639-C04-02

Additional details