From Multiple Views to Textured 3D Meshes: A GPU-Powered Approach

A GPU-Accelerated GVF Snake Algorithm

“…Relevant to individual vision algorithms, Tzevanidis [7] implemented a complete real-time 3D reconstruction system based on visual hull and executed on GPU, including image segmentation, visual hull calculation, Marching Cube (MC), surface smoothness and texture mapping, etc. Their work proved that almost all of the core algorithms in a 3D reconstruction system could be transplanted to the GPU.…”

Multicamera tracking of multiple humans based on colored visual hulls

“…The proposed work capitalizes on computational efficiency in 3D reconstruction to facilitate person tracking, because as person motion becomes more densely sampled in time, it also becomes less ambiguous to track. Thereby, we adopt the work in [20] which provides person tracking based on efficient volumetric reconstruction achieved with the software platform presented in [21,18]. In the current work, we additionally employ color information to extend the above approach in order to disambiguate challenging tracking cases.…”

“…One computer with programmable GPU hardware (with the option of sharing computational effort in more computers, if required [18]), is responsible for image acquisition, processing and extraction of a spatial representation of the persons in the room. In a typical setup, the cameras are placed evenly and high on the walls of the room, to overlook the scene.…”

“…Foreground detection is performed based on a GPU implementation [18] of [23], which parallelizes computation at a pixel level. The outcome of foreground detection is binary image B i .…”

“…The surface of the visual hull is obtained as the 0-isosurface in V [22]. The isosurface is computed by a parallel implementation [18] of the "Marching Cubes" algorithm [11]. The 0-isosurface is encoded as a mesh M of triangles.…”