Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : June 01, 2022 - June 04, 2022
In this paper, we implement a novel method to estimate road surface shape and object height using a fisheye stereo camera on GPU for acceleration. Our method consists of three parts, preprocessing, road surface shape estimation and object clustering. Among them, the road surface shape estimation algorithm takes 20 seconds to process because the algorithm is recursive. This is not feasible for real application. Therefore, we change the algorithm to be able to be processed in parallel. We conducted experiment to confirm the difference between the accuracy and the processing time between the previous CPU implementation and the proposed GPU implementation of the approach. It was found that the proposed GPU implementation resulted in 8-times faster processing with no loss of accuracy.
