Real-time Multicopter Detection Using Pixel-level Digital Filters for Frame-Interpolated High-frame-rate Images

摘要

This study proposes a fast algorithm for vision-based vibration source localization that can detect vibration sources at hundreds of Hertz by inspecting time-varying brightness signals at each pixel in high frame rate (HFR) images. Our algorithm can significantly reduce the computational complexity of pixel-level digital filters for vibration source localization by virtually adjusting the sampling rate to twice the vibration frequency of a target object to be tracked using downsampled HFR images with frame interpolation. We confirmed that a target object vibrating at 200 Hz or less can be localized in real time with pixel-level digital filters using$pmb{512}imes pmb{512}$images at 1000 fps without degrading the filter properties when our algorithm was implemented by software on a personal computer. The effectiveness of our algorithm was demonstrated by showing the experimental results for a flying multicopter with four dual-blade propellers rotating at 95 rps in an outdoor location.