Background Subtraction Algorithm Based On Multi-Camera Corporation With Application On In-door Tracking

There has been many research on real-time background subtractionalgorithm. A well-known assertion says no algorithm meets all conditions.Background subtraction is a fundamental step for further image informationanalysis. A multi-camera co-operate background subtraction algorithm,designed for in-door human tracking, is able to merge the calibration and colorinformation of cameras, and tends to show results with more accuracy thosefrom single camera algorithms..We apply the multi-camera co-operate background subtraction algorithm tothe task of in-door human tracking, and designed a suitable particle filtertracking kernel. A special tracking strategy with concern about human motionpattern is presented.The major contributions of this paper are: First, we introduce a logarithmiclumination distance based on the observation that the lumination threshold forbackground subtraction is almost constant in logarithmic scale under differentlight condition. Experiments show its tolerance with a variety of luminationvalues. Second, a background generation algorithm based on muti-cameracorporation. Experiments show that this algorithm produces stable and reliablebackground in in-door environment. Third, a slightly improvement on particletracking algorithm with a newly designed particle kernel and a search strategyconsidering patterns of human motion.Experiments are made comparing the background substraction performanceof some single-camera algorithms and the method we presented. The results arepromising. The improvement on particle tracking is also testified.