ABSTRACT
Data parallel Graphics processing units (GPU) computing allows for real-time three-dimensional (3D) mapping of scenes from depth cameras such as the Kinect sensor. Low-cost depth cameras using structured light or time of flight methods have become commonplace. The sparse octree used to represent a reconstructed 3D map will quickly grow too large to fit entirely in GPU memory. OctoMap is a probabilistic framework where the log-odds of occupancy are stored in an octree data structure. The sparse octree structure overcomes the scalability limitations of a dense voxel grid by leaving free space unallocated in memory. Augmented reality (AR) is a field that lives on the boundary between computer graphics and vision to create experiences that blend artificial graphics with the real world. AR systems today typically render virtual objects in front of a raw depth camera frame. The hard part is computing the error gradient fast enough to keep up with the camera’s motion for the solution to converge.
