Moving target detection in complex scenes based on spatio-temporal domain analysis

This paper proposes a method to suppress the false positives of dynamic background based on Mixture of Gaussians (MOG) background modeling in space domain and time domain. In space domain, MRF and MOG model are used to calculate the class-prior probability and class-conditional-probability of the pixel. Foreground segmentation is completed by combine of the class-prior probability and class-conditional-probability of the pixel, and most false positives of small size are suppressed. In time domain, false positives are suppressed on three target characteristics: motion constancy, motion saliency and area stabilization. The results of experiment show that the false positives of dynamic background could be suppressed to a great extent while good detecting accuracy is ensured.