Multi-Channel Adaptive Mixture Background Model for Real-time Tracking

Extracting the segmentation of moving regions in image sequences in realtime is a preliminary stage for many computer vision tasks such as video camera surveillance and human-machine interface communication. A typical method for real-time moving region detection is background subtraction. The first step is to construct a background model. Numerous background models have been introduced to solve this problem for different requirements. The most efficient one is Gaussian mixture model, but it still exists some problems. This paper discusses old modeling methods and proposes a new method base on chromatic channels to construct a background model. By reviewing the existing background update equations, we find some inappropriate points and take some modifications. This makes the method adaptive more accurately to a changing environment without consuming too much time. The result reveals that our improved method is efficient and well-performed.