A Multi-view Camera System for the Generation of Real-Time Occlusion-Free Scene Video

Introduction Multi camera setups are bound to have occlusions. It would be very useful if we resolve these occlusions. The problem is to find an approach to resolve these occlusions The proposed approach actually works very well. Below is an experiment of a man passing behind an occlusion. For indoor setups, we use structured light to find occluded areas in all views. Structured light source is placed as close as possible to the main view. Later, u-v code maps p pp in real-time. p , p are used to find the occlusions. This preprocess makes the run-time phase faster and helps to find the perspective transformations between views. For outdoor setups we make a cross-check to determine whether an object is occluded or not. (a) A scene containing a wall and a dynamic sphere. (b) View of the scene from main camera. (c) View of the scene from second camera. (d) " Occlusion-free " view of th f th i i (a) Main view of the scene with two occlusions. (b) Helper view of the scene. (c) low order u codes for the main view. (d) low order u codes for helper view. (e) Occluded areas the scene for the main camera view. Such a system would be very useful for both automated and manual processing.