The explosion in available sensory data has been one of the greatest technological developments of the last 25 years. Particularly, the availability of CMOS image sensors of ever greater quality, together with exponentially growing computing power, has enabled numerous new visual monitoring applications in commercial, industrial, scientific and military fields.
We aim to develop new analysis methods for applications requiring real-time image content analysis with rapid reaction to visual data. This includes e.g. high-speed visual control of industrial processes, such as laser welding and 3D manufacturing. We apply existing and upcoming parallel processing platforms, such as sensor-level processing with custom ASICS and FPGA-based embedded platforms, with dedicated parallel algorithms, to achieve sufficient performance for very demanding real-time applications.
In our approach the whole image analysis flow is optimized from sensor-level capture and low-level early image segmentation to higher level image content understanding, in order to maintain the highest possible level of parallelism in all stages of processing. To achieve this, our research also takes advantage of new sensor-processor and cognitive computing technologies developed at DFT/Microelectronics.