How can we create a comprehensive and dynamic representation of a video scene? This paper presents a novel framework for video mosaicing that separates background, foreground, and object motions. It relies on the Dominant Motion Assumption (DMA), which assumes that the background exhibits parametric motion and occupies most of the scene. A robust optical flow-based method extracts moving parts using mixtures of Gaussians. Summarization of background, summarization of foreground, and trajectories of objects is a benefit. Trajectories of objects in the scene are summarized. This mosaicing method offers a powerful tool for video analysis and summarization. The method could have applications in surveillance, video editing, and computer vision. The method provides a computationally efficient way to create detailed representations of dynamic scenes.
This paper is appropriate for the International Journal of Pattern Recognition and Artificial Intelligence because it describes a novel method for video mosaicing using robust pattern recognition techniques. The integration of Bayesian methods further aligns with the journal's focus on intelligent algorithms for image and video processing.