A statistical parameter learning method for cast shadow model

In special video surveillance environment, the intensity variance of cast shadow of independent moving objects has its own special statistical model. This paper proposes statistical parameter estimation method for cast shadow of moving objects, based on statistical correlation in the situation of stationary cameras. In view of pixels belonging to moving shadow show stably statistical characteristics, while that belongs to different moving targets have weak correlation among them, we obtain a stably statistical distribution of shadows by a correlation calculation of histograms from many detected moving regions. It could give a credible partition between moving targets and moving shadows. Simple shadow detection method based on our statistical model can be used to detect cast shadow of moving objects. Experimental results demonstrate that our technique can detect moving cast shadows robustly in an efficient and simple way.