Embedded Vision Systems – Paving the Way to More, Faster and Cheaper Machine VisionApril 7, 2017
The application of imaging in industrial manufacturing, medical devices, traffic, transportation, logistics, life sciences and research is often a challenge, despite the great technical advances over the last 10 years. Multiple hardware components, such as optics, cameras, cabling, data acquisition, processing and storage units need to be considered, and need to comply with high data rates and computationally intensive algorithms. Thus, classic vision systems are often based on high-performance PCs.
When just a few further requirements come into play, however, embedded vision systems become the best, if not the only solution:
Constraints in space, temperature, or mechanical robustness
Embedded vision systems allow for much higher spatial integration of data acquisition, processing, storage and output components. Further, unlike PC systems designed for IT or consumers, embedded systems can be optimized to robustness against extreme temperatures, vibrations and mechanical shocks by careful selection of parts, connectors, PCBs and manufacturing processes. Active Silicon has more than 25 years of experience in the integration of electronic components for challenging applications in defense, marine, space, medical, and automotive.
Embedded systems also have the advantage that components can be specifically selected with long-time availability in mind. The careful supply-chain management at Active Silicon Silicon assures that embedded systems typically retain the same form, fit and function for at least 10 years.
Processing speed and power consumption
Image processing is characterized by high data volumes derived from each image represented by large 2D-pixel matrices with millions to tens of millions of entries – and with each pixel carrying up to 16 bits of greyscale or 24 bits or more of color information. Additionally, cameras with high frame rates deliver several dozen if not hundreds or even thousands of images per second.
Even high-end CPUs are not capable of applying even moderately complex algorithms on this amount of data. Modern GPUs are a solution, yet power hungry and costly. Instead, the latest Field-Programmable Gateway Arrays (FPGAs) can be optimized for parallel signal processing and thus are ideally suited for images and video streams. Despite comparably low clock rates and very low power consumption, they outperform general purpose GPUs and CPUs depending on the algorithm. Latest integrated architectures comprising CPU and FPGA provide efficient implementation options for algorithms with parallel and serial processing needs.
As embedded vision systems are customized to the requirements of the specific application, the selection of the right components and the optimization of production processes can reduce manufacturing costs considerably.
Active Silicon’s ready-to-use embedded solutions and modular system designs allow for short time to market and significantly lower development costs.