Abstract
A detection algorithm of moving objects which is robust and effective is proposed for the problem of the chaos and variety scenes. The background model based color quickly is build during the nonrestraint learning. According to the background of the chaos, it updates the model. However, shadow is also detected as moving object because of the same characters between them, so the algorithm of shadow removal based on its characteristic of chroma, lightness and crossover entropy was presented too. Finally, the algorithm was simulated in real-time and experimental effect, and achieved better results.