We present a simple and powerful method for planar object detection which is robust against changes in camera positions using image warping. Applying stereo camera, we efficiently reconstruct the sparse 3D scene and find the dominant plane. We then place a virtual camera facing directly at the selected plane from the predefined distance and angle so that the plane of interest appears in the intended size without any affine distortion due to a slanted view point. The exact 3D location of any point on the surface can be recovered by oriented chamfer matching using a single template and inverse warping. This algorithm shows a good performance on indoor robot vision application which often requires detecting planar objects such as locating an elevator button or recognizing signs on the wall.
|Jiyoung Jung, Yekeun Jeong, Joon-Young Lee, Hanbyul Joo, In So Kweon
|The 7th International Conference on Ubiquitou s Robots and Ambient Intelligence (URAI)