Multi-stage Moving Object Recognition Based on Fuzzy Integral

A multi-stage objects recognition process based on biomimetic pattern recognition (BPR) and Choquet integral (CI) is proposed to detect and classify the moving objects in video sequence in the intersections. It is difficult to distinguish motorcycle from pedestrians when occlusions happen. In order to solve the problem, BPR is first used to classify the Zernike moments extracted, and CI is then adopted for multi-features fusion based on the output of BPR, the area and the velocity to improve the accuracy. An Experimental example is proposed to test the efficiency of the approach presented.