Abstract
We present a mobile robot that can actively localize a sound source behind an object, approach it by avoiding obstacles on its path, and discriminate whether the source is a human or not. The robot has a unit of three microphones for ears and a CCD camera for an eye. The direction is estimated based on time differences among signal onsets received by the microphones, located at each vertex of an equilateral triangle. In a complex scene where a source is not visible from the robot, it estimates tentative goal direction based on diffracted sound at the edge of an object occluding a sound source. Repeating source localization at different positions, the robot find the source and discriminate it by block matching an image captured by the camera mounted on the robot with template image memorized in advance. We have implemented these methods and obtained satisfactory results demonstrating the effectiveness of the methods. We also describe a method for estimating sound source distance based on observations at different positions.
