3D image acquisition and processing with high continuous data throughput for human-machine-interaction and adaptive manufacturing
Many applications in industrial environment are able to detect structures in measurement volumes from macroscopic to microscopic range. One way to process the resulting image data and to calculate three-dimensional (3D) images is the use of active stereo vision technology. In this context, one of the main challenges is to deal with the permanently increasing amount of data. This paper aims to describes methods for handling the required data throughput for 3D image acquisition in active stereo vision systems. Thus, the main focus is on implementing the steps of the image processing chain on re-configurable hardware. Among other things, this includes the pre-processing step with the correction of distortion and rectification of incoming image data. Therefore, the approach uses the offline pre-calculation of rectification maps. Furthermore, with the aid of the rectified maps, each image is directly rectified during the image acquisition. Afterwards, an FPGA and GPU-based approach is selected for optimal performance of stereo matching and 3D point calculation.