Our Embedded Firmware team is specialized in the design and development of state-of-the-art signal and image processing algorithms for highly constrained embedded devices. Truly integrated firmware design together with our custom ASIC controllers and digital Hardware accelerators enables us to provide highly efficient embedded and edge computing solutions for a wide variety of Image and Signal applications.
We have experience on-device implementation of adaptive filtering, denoising, whitening, and other models-based filtering techniques for Image/signal pre-processing to further enhance the quality of sensor readouts in presence of strong interferences which are typical of industrial or medical applications.
Our image processing controllers include robust image segmentation, active area, and shape detection along with continuous tracking of up to 128 shapes simultaneously. Tracking and trajectory predictions based on k-means clustering and predictive filters enable our image processing controller to individually detect, assign and track all the shapes reliably. We use fast object detection and region-based classifications using convolutional filters to reliably classify the true events in very harsh environments.
Our latest innovative algorithms implementation uses recurrent neural net (RNN) for adjacent channels of active pixels and enables our platforms to reach best-in-class performance in precise subpixel localization typically for a tip on a large pixel grid (< 0.45mm for a 1cm grid) even for very large pixel grids unparallel to none. We use Kalman particle filters for tracking and model-based trajectory predictions to reach an astonishing naturally feeling movements.