The Research For Humanoid Robot Acoustic Tracting System

The audition means much for robot to get information from environment and control its behavior. The localization and tracking of object are widely used in the fields of military, meet record, and so on.Mobile robot can be steered to the target by following the Sound source direction; humanoid robot's audition is an important way to intercommunicate between robot and human. Sound source localization and tracking often greatly supports to demonstrate the functions above, such as the fusion of multi-sensor, recognition of speech, etc.The main difficulty of sound source localization based on robot lies in the elimination of noise, especially the noise submitted by itself, such as by the operation of electromotor, the friction of joint and the vibration of weakly-rigid structure; and for multi-target tracking system, the difficulty is to conform the identity of the same target, especially when several sound object are in the same area.Hence, this paper presents the sharp beamforming arithmetic which is based on the classical beamforming method and contains new way to get sharper cross-relation and beamforming peak. In addition, the system applies Unscented Kalman Filter to forecast the contrail of tracked objects, which helps to realize the conformation of the target's identity.According to speed of development and function, the system use three microphone on the robot head;The experiment shows that after brief prepare when there is no target of sound, the audition system behaves robustly and precisely in low signal-to-noise condition and reacts more intelligently to multi-target.