Rotational motion estimation using motion blur of AMCW LiDAR Pohang University of Science and Technology

Abstract

In this research, proof-of-concept research on rotational motion estimation using only the motion blur of AMCW LiDAR is done. A neural network model is proposed to predict angular velocity in both pan and tilt directions, utilizing a dataset of AMCW LiDAR images with motion blur were specifically created for this purpose. To enhance performance, dataset post-processing techniques, such as single-object-cropping and uneven rotational velocity distribution, are introduced. In a tilt direction, the MAE is measured to be 3.20 deg/s, with a standard devi- ation of 2.82 deg/s, while in a pan direction, the MAE is 4.99 deg/s, with a standard deviation of 3.92 deg/s. The result suggests that the motion blur of AMCW LiDAR is not merely an artifact to be eliminated but contains valuable information. It also shows the potential of extracting meaningful information from motion blur for practi- cal usage, such as motion prediction tasks.