抄録
This paper discusses environment recognition method from a camera system with translational and rotational mechanism. The system is expected to show better estimation accuracy even when the camera's baseline length is small, thus it can be mounted on a small exploration rover. In our previous work, geometric parameters of the camera system have been identified from camera images of artificial objects with sharp boundaries in indoor experiments. This paper applies the proposed system to natural objects, e.g. stones and sand, and verifies the estimation performance for outdoor applications. The effect of changing light condition is also experimentally evaluated. Finally, we propose a procedure to reduce the expected area that feature points exist.
