On the Use of Calibration for Pedestrian Detection in On-Board Vehicular Cameras

This paper presents a new approach for pedestrian detection in the context of Driver Assistance Systems (DAS). Given a camera with known intrinsic parameters, a flexible online calibration scheme that explores the expected road geometry is used to obtain the extrinsic parameters. With the full camera parameters, the expected geometry and size of a standing person is used to customize a baseline pedestrian detector based on sliding windows and multiple scales. Our experimental results show that the proposed approach allows the use of detachable cameras in the context of DAS, improving the accuracy of the baseline pedestrian detector. Furthermore, the flexible calibration scheme allows to estimate the distance from detected pedestrians to the camera using detachable cameras, opposed to the fixed onboard cameras in commercial vehicles that support vision-based DAS.