As camera frame rates increase, true real-time vision-controlled systems will become cost-effectively available. Visual inspection loops of 2D image computing at 1,000Hz will be reality. We design and implement FPGA-based vision-in-the-loop solutions for real-time, precision quality inspection in a control & feedback loop.

Modelling, manufacturing, and measuring freeform optics is maturing. Cost-effective aspheres, rotational invariant discontinuous surfaces, structured surfaces, arrays of lenses or pyramids can be introduced in optical systems. This allows optical engineers more design freedom in areas such as semiconductor development, consumer electronics (HMD, mobile phone cameras, etc.), general lighting and automotive applications (LED) and medical e.g., personalized ophthalmology.

Computation power is increasing rapidly and allows deep learning and neural network combinations for support tasks in recognition. Custom-made neural networks with proper data pre-processing will have a clear advantage over the more general neural networks without proper pre-processing. We design such custom-made inspection algorithms based on our concepts. In our concepts, the customer can create and train the network themselves without having knowledge of deep learning or neural networks. We also use synthetic data to train our neural networks. This helps to overcome the limited available training data set, and is useful when multiple defects need to be recognized.