Trade-off Between Resolution and Frame Rate for Visual Tracking of Mini-robots on Planar Surfaces

摘要

Accurate and fast visual localization is required in many applications of mini-robotics. Achieving both fast and accurate algorithms is difficult due to computational and acquisition issues. Higher image resolution typically implies lower frame rates which results in blurred images of fast moving robots. We aim to achieve as high frame rate as possible with acceptable accuracy of localization. Specifically, we propose a two stage localization algorithm based on fast block matching for fast localization followed by optical flow correction for subpixel accuracy of localization. We show in experimental evaluation that the difference in localization using an image of the same scene in resolution 1000 × 1000 px and 200 × 200 px is 0.2 px of the higher resolution. Computational cost of the lower resolution is 16 times lower than that for the high resolution. This allows us to achieve high accuracy localization at higher frame rates which significantly reduces the blur artefacts caused by low frame rates. Experimental results are shown to demonstrate the precision and speed of the proposed algorithm in the tracking of a robot on a recorded video.