| PUBLICATION |
| Full record |
|Title:||Real-Time Voxel-Based Visual Hull Reconstruction|
|Where:||XXV Conference on Design of Circuits and Integrated Systems, DCIS'10|
Jesús Miguel Pérez
Pablo Pedro Sánchez
Design and verification of HW/SW embedded systems
|Abstract:||High-definition 3D video for real-time applications is one of the features that the next generation of telecommunication systems is exploring. Some of these systems include Visual Hull (VH) reconstruction based on “shape from silhouette” (SfS) methods.|
From the implementation point for view, the most restrictive limitations of all the previously published methods based on SfS are the processing speed and the memory and resource consumption. In real-time video applications, the latency (algorithm execution time) must allow the processing of 25/30 frames per second. Moreover, there are also strict limitations in the maximum number of memory accesses and parallel operations that can be implemented in a real hardware platform.
In this paper we present an architecture for SfS reconstruction which outperforms the processing speed of the latest publications, allowing real-time, high-definition VH implementations. The architecture is modular and the number of cameras can be increased as required without processing speed penalty. We also present some algorithmic improvements to reconstruct VH using a voxel-based approach, reducing the resource consumption. Finally, we illustrate the advantages of the
proposal with an FPGA-based implementation. The new approach allows a 256x256x128 reconstruction volume to be obtained in only 33.55 milliseconds using a single Virtex 5VLX330.