Inline Computational Imaging (ICI) is a novel single sensor technology, capable of simultaneous 2D and 3D inline inspection invented by AIT Austrian Institute of Technology GmbH. It combines the advantages of light field imaging and photometric stereo into one compact solution. ICI technology is a new type of image acquisition system, combined with smart algorithms for high resolution and high speed 2D and 3D inspection (standard setup: 20µm/pixel optical resolution and maximum acquisition speed of 500mm/s). The underlying ICI software processes area image data and produces 3D point cloud data.
To render ICI point cloud data in the browser, AIT evaluated Potree as an appropriate solution. Potree streams and renders point cloud data pre-processed with PotreeConverter proposed in “Fast Out-of-Core Octree Generation for Massive Point Clouds”. This software converts point clouds to a hierarchically Level-of-detail (LOD) structure, required for web based consumption.
As the complete ICI 3D-reconstruction pipeline is implemented in CUDA and runs entirely on the GPU, AIT wants to perform the point cloud to octree conversion on the GPU too. This master’s thesis aims to implement the complete octree generation process for Potree on the GPU using CUDA. Beside the performance improvements, this approach enables a direct compatibility with the point cloud results from the ICI pipeline. To minimize Aliasing artifacts during rendering, the octree will be implemented using a replacement scheme instead of an additive scheme.