Real Time Analysis on Bus Passenger for Unmanned Door Operation using Overhead Fisheye Cameras

In this study, we propose a robust method for ensuring safe unmanned bus door operation using commonly available surveillance cameras. Bus door operation is a critical function of bus services because passengers can be injured unless they work appropriately. Our method leverages only image processing technologies and is composed of an object detector, an optical flow analyzer, and a pose estimator. We carefully combined the modules so that they judged the readiness of the door operation accurately and in real time. We confirmed that our proposed method can ensure safe bus door operation using only two industry fisheye surveillance cameras installed on a modified Toyota Coaster microbus. Our system runs at 10 fps using a commonly available GPU desktop.