振動ベースド画像特徴量を用いた高速ターゲットトラッキング

概要

論文の詳細を見る
In this paper, we propose a novel concept for the real-time target tracking of vibrating objects to extract vibration image regions using band-pass digital filters at all the pixels in an image, assuming that significant periodic changes in image intensities exist at pixels around vibrating objects. By combining vibration-based image processing with color feature extraction as a masking process, our algorithm can prevent the latency effect on a digital filter, which may degrade its tracking performance. Band-pass digital filters at a pixel level are actually implemented on a 1000-fps real-time vision system that can measure vibration distribution over a wide range of frequencies. By using our target tracking system, we can continuosly track a vibrating object at the center of the camera view by controlling a 2-axis active vision system with vibration-based image features.