Video stabilization taken with a snake robot

Abstract

We design a digital image stabilization algorithm, to be applied in videos captured with a camera installed in the initial module of a snake robot. Stages for digital image stabilization were: motion estimation and motion compensation. In the case of motion estimation, we selected and evaluated two techniques: Pyramidal Lucas-Kanade optical flow and integral projections, using SAD and SSD as methods for comparison between the signatures. The system was implemented in OpenCV and their outputs were a video achieved the final compensation of the input video, and statistical measures of mean and standard deviation of the motion estimation between frames. The final algorithm can be implemented in real time due to its low computational cost and can be used as a pre-procesing method to acquired visual information from snake robot.