High-speed Traget Tracking with Vibration-based Image Features

Abstract

This paper describes a real-time target tracking system for vibrating objects that can utilize temporal frequency properties in brightness as image features for target tracking by executing frequency filtering for brightness at each pixel in images, assuming that there are strong temporal changes in brightness on many pixels around vibrating objects according to their vibration frequencies. By implementing a target tracking algorithm based on pixel-level frequency filtering on a real-time HFR vision system that can measure vibration distribution at several dozens or hundreds of Hz, we develop a 1000 fps target tracking system that can always track vibrating objects in the center of its camera view by controlling a 2-DOF active vision with visual information. Its effectiveness is verified by demonstrating several experiment results of vibrating objects at a frequency of several dozens of Hz, which are too fast to see for human eyes.