With the rising importance of automation systems, the demands on computational requirements increase as well. Industrial automation systems not only have to perform complex calculations on huge amounts of data in real-time, they also have to be reliable, which is essential for industry. Combining all requirements, it has increasingly become impossible to be achieved by a single PC. Fortunately, there is High Performance Computing which provides various concepts of e. g. distributing algorithms to a cluster of PCs. Although HPC is common in scientific research, it is yet rarely found in industrial applications even though HPC can bring significant improvements. For this reason, we want to demonstrate how a HPC communication model with distribution and self-balancing mechanism can be applied to a concrete application of industry in practice.

In recent years, a large number of multitouch sensor concepts have been presented. Particularly optical sensors are highly popular due to their versatility. However, especially camera-based systems often require a significant amount of space behind the screen and are not well suited to flatscreen-based setups. While integrated sensors for flatscreens have already been presented, they are mostly complex, expensive or both.

In this paper a novel implementation of the saliency map model on a multi-GPU platform using CUDA technology is presented. The saliency map model is a well-known computational model for bottom-up attention selection and serves as a basis of many attention control strategies of cognitive vision systems. A real-time implementation is the prerequisite of an application of bottom-up attention on mobile robots and vehicles. Parallel computation on graphics processing unit (GPU) provides an excellent solution for this kind of compute-intensive image processing. Running on 1 to 4 NVIDIA GeForce 8800 (GTX) graphics cards a frame rate of 313 fps at resolution of 640 times 480 is achieved, which is approximately 8.5 times faster than the standard implementations on CPUs. The implementation is also evaluated using a high-speed camera at 200 Hz. Using two GPUs only 2 ms extra computational time for the saliency map generation in addition to the camera capture time is required for images of 640 times 480 pixels.