An Adaptive Kernel Density Estimation for Motion Detection

This paper proposes a method of adaptive kernel density estimation (KDE) for motion detection. The method selects an adaptive threshold by analyzing probability histogram, which is suitable for different scenes and different moving objects. Then a mechanism of updating background using probability is also provided. It can get relative good background and is useful for motion detection. Moreover it can solve deadlock situations in updating background model. Some improvements are proposed to reduce computational cost for real-time applications. Experiments show the method is effective and efficient.