An Extraction Method of Moving Object Based on Visual Cognition Mechanism

Reichardt motion detector is the most basic mechanism of motion perception of biological vision system. But because of the aperture problem and the dependence of scene pattern, the detected motion information is always ambiguity. However, the vision cognition mechanism shows that: the motion perception of vision system doesn´t only detect motion velocity, but also supplies the temporal feature to assist object recognition. Thus this paper puts emphasis on the contribution of temporal features to the moving object recognition and extraction, rather than the accuracy of motion estimation. A spatial-temporal iterative feed forward method was proposed to identify the moving objects. The common motion information is propagated along the similar spatial features; the temporal synchrony binds spatial features of the same object. By the interactive process, the moving object will be recognized and extracted by the bound spatial feature, and the ambiguity motion information will be corrected. The proposed method can overcome error motion information caused by noise, camera-self oscillation and poor luminance.