Embedded Implementation of a Real-Time Motion Estimation Method in Video Sequences

摘要

This paper presents an embedded implementation of a hardware-efficient method for motion information extraction from a video signal. The high computing cost of gradient-based methods makes their implementation in compact devices for real-time applications challenging. We present our achievements in implementing a novel, simplified version of a gradient-based method, using an FPGA circuit-based system for on-line Sobel edge detection and edge displacement-based optical flow computation. The image data is acquired from a dedicated embedded camera transmitted to the reconfigurable platform. The design and implementation steps of the devised method are presented, followed by the description of the system's setup and measurement results. The yielded results are compared to similar implementations. These results lead to possibilities in using the developed system for the enhancement of indoor mobile robot localization.