Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : June 28, 2023 - July 01, 2023
The purpose of this research is to improve the accuracy of pedestrian crossing detection for the development of sidewalk mobile robots. Specifically, we used YOLOv7, an object detection method using a deep neural network, to target pedestrian crossings with horizontal white lines on black roads. Then, three patterns were compared: detecting the entire pedestrian crossing, detecting a part where the pedestrian crossing lines are arranged in black and white order, and detecting a part where the pedestrian crossing lines are arranged in black and white order. In conclusion, we were able to detect the entire pedestrian crossing with high accuracy. However, it was found that there are cases where a horizontal stripe pattern like a crosswalk is mistakenly detected as a pedestrian crossing.
