An Energy-Efficient Reconfigurable AI-Based Object Detection and Tracking Processor Supporting Online Object Learning

Abstract

This letter presents an energy-efficient reconfigurable AI-based object detection and tracking processor for smart drone/robot applications. Several techniques have been proposed to achieve high energy efficiency while supporting flexible object detection and tracking tasks with online object learning, including a reconfigurable object detection and tracking architecture with reconfigurable neural network (NN) engine, an online object learning architecture with shared NN inference and learning engine and automatic label generation engine, and a layer- and stride-aware NN computing technique. Compared with several state-of-the-art designs, the proposed design achieves better energy efficiency (2.13 mJ/frame), while supporting flexible object detection and tracking tasks with online object learning.