Dense 3D Modeling Using Consistency-Based Feature Point Evolution

To render scenes by camera-based ambient intelligence, we propose a dense 3D modeling algorithm using feature points that dynamically appear and disappear over time; that is, we reconstruct a 3D model of the scene with a dense feature point set that “evolves” over time. As the scene’s appearance changes due to camera movements, some existing feature points dynamically disappear while some new feature points dynamically appear relative to the camera. The newly generated feature points’ 3D positions are initialized using nearby existing feature points’ positions depending on their consistency including the distances in 2D image, stability and history. Our feature evolution, when incorporated into standard tracking and 3D reconstruction algorithms, provides more robust and denser 3D meshes. Consequently, the resulting 3D meshes and textures render novel 3D images better than meshes and textures produced using standard stereo techniques.