Object Detection in Curved Mirror with Multi-Cameras from Single Viewpoint Video

Compared to other countries, Japan has many T-intersections and intersections without traffic lights. Accidents occurred on T-intersections becomes a significant problem because of the large ratio in traffic accidents. For this reason, a large number of curved mirrors are set near dangerous areas. However, due to the limited visibility of the human eye, the recognition of curved mirror images using an on-vehicle camera is effective. In addition, for the realization of automatic driving on narrow roads in Japan, highly accurate recognition technology for curved mirror images is essential especially at a long distance to the curved mirror. In this work, we propose a system for detecting approaching objects in a curved mirror video using an inexpensive single view multi-camera.