RESEARCHES

We propose a high-frame-rate (HFR) vision based surveillance solution for real-time multicopter detecting and tracking. The proposed vision-based vibration source localization algorithm based on the vibration-based image features, which are obtained by detecting periodic changes in image intensities at pixels around rotating propellers of multicopter, vibrating at audio-level frequency (several hundred hertz) by using pixel-level band-pass digital filters. Clearly different from conventional appearance-based object tracking with spatial pattern recognition, our algorithm is robust with various factors of appearance changes, such as brightness, defocus blur, apparent scale, pose variation, complicated background, partial occlusion and motion blur. Therefore, is robustly applicable in real scenario.
The algorithm was implemented on a high-speed vision platform with a 2-DOF active vision system, which is controlled by visual feedback at 1000 fps that always track the target multicopter at right center of the camera view. The proposed HFR-vision-based method with pixel-level digital filters enables vibration source localization with sub-degree-level angular directivity in real time.