A New Concept of Pose Estimation of Arbitrary 3D Object without Prerequisite Knowledge: Projection-based 3D Perception

Generally speaking, the tasks that utilize robot vision such as visual servoing or pose estimation towards a solid object require a prerequisite condition. Because the system can only check out objects with predefined features or models, the requirement of a priori knowledge of the target object seems to have been a hindrance for tasks applied for the arbitrary target. For this reason, a new approach named: Pb3DP (Projection-based 3D perception) is proposed in this paper. The concept of Pb3DP employs stereo-vision effectively, that is (1) use the 2D image of the target object in the left camera as the 2D model, then (2) inversely project the model into 3D space with an assumed pose, and then (3) the projected 2D model in space onto right camera’s image again. The projection (2) and (3) are calculated by computer. Here, (4) if the 2D model reprojected by the computer coincides with the real 3D target object projected naturally by camera’s function, then the assumed pose could be thought to present the real target’s pose. In this paper, we apply Pb3DP to estimate a solid target object’s pose in real-time, by showing the experimental result, the effectiveness of the proposed method will be discussed.