Infinite 3D Modelling Volumes

Recursive Total Variation Filtering Based 3D Fusion

“…Early real-time dense approaches (Stühmer et al, 2010;Newcombe et al, 2011) were able to estimate the image depth from monocular input, but their use of a dense voxel grid limited reconstruction to small volumes and to powerful GPU devices due to high memory and computation requirements. Also suffering from the scalability issue, KinectFusion (Izadi et al, 2011;Funk and Börner, 2016) directly sensed depth using active sensors and fused the high quality depth measurements over time to surfaces. The scalability issue has been resolved by voxel hierarchy approaches (Chen et al, 2013) or direct voxel block hashing (Kähler et al, 2015) to avoid storing unnecessary data for free space allowing scaling to unbounded scenes.…”

“…This work addresses both critical challenges identified from the literature, which are: i) The reconstruction of unbounded environments. This is addressed by the method of (Funk and Börner, 2016). ii) The 3D fusion of error prone depth data into a volumetric 3D space, which is the main aspect of this work.…”

“…In actual implementation the algorithm was implemented using threads to maximize the efficiency and utilize the latest CPU architecture. We start our fusion application by instantiating an object of data structure proposed by (Funk and Börner, 2016)…”

“…al, 2008) wearable and/or assistive technology (goo, 2014;Hicks et al, 2013), and change detection (Taneja et al, 2013). Recently a few methods have been proposed to perform incremental reconstruction (fusion) of the observed geometries from RGB-D measurements (Funk and Börner, 2016;Kähler et al, 2015;Steinbruecker et al, 2014).…”