Deep Learning and Embedded Vision
By using smaller yet higher-performance networks and direct image transfer from the camera to image processing, CNNs are predestined for embedded Vision applications. They can run on frame grabber FPGAs as well as on VisualApplets-compatible cameras and vision sensors. Particularly when using the Industry 4.0 decentralized computing approach, there arises a need for embedded Vision with deep learning. Small image processing units or even intelligent cameras are already taking over demanding subtasks even now.
Since most embedded devices are equipped with an FPGA, they possess sufficient performance for more complex neural networks. In comparison to GPUs, FPGAs are especially energy efficient and good for embedded and industrial applications with hard real-time conditions, such as inline inspection, robotics and pick and place, and cognitive systems, as well as human-machine interaction (HMI). Other applications with very low error rates are found for example in the fields of quality assurance, medical technology, drones, and automotive (autonomous driving).