Moving object detection of dynamic scenes using spatio-temporal context and background modeling

Within the field of automated video analysis, detection of moving objects remains a challenging task due to the presence of dynamic background and camera motion. Dynamic scenes contain some moving objects such as trees jiggling slightly and water flowing irregularly. In this paper, we present an algorithm to address the problem of dynamic background, which employs spatio-temporal context and background modeling according to Bayes theorem. Spatial context refers to connections of pixels exist almost everywhere while keeping interrupted at boundaries between foreground and background. We use spatial context to eliminate noise points and obtain continuous foreground region. Temporal context interacts with mixture background model, which alleviates spurious detection of dynamic scenes. Object detection is finally carried out by minimizing the energy function of formulation in Markov Random Field. Employing spatio-temporal context helps to sustain high levels of detection accuracy. The efficiency of our algorithm is demonstrated by experiments performed on a variety of challenging video sequences.