NCN Preludium

Acceleration of processing event-based visual data with the use of heterogeneous, reprogrammable computing devices

Akceleracja przetwarzania zdarzeniowych danych wizyjnych z wykorzystaniem heterogenicznych, reprogramowalnych układów obliczeniowych

Principal Investigator: MSc Marcin Kowalczyk

Project Duration: 28 January 2021 - 28 July 2026

Budget: 185440 PLN

The aim of the proposed basic research is to design innovative methods of processing video data using modern neuromorphic sensors and heterogeneous MPSoC (Multiprocessor-System-on-Chip) devices. This should allow to increase the effectiveness of vision systems, reduce power consumption, and lower the latency of data processing. Event cameras are neuromorphic sensors that are inspired by the way the human visual organ works. Contrary to traditional cameras, the output does not provide whole frames, but only information about the pixels for which the brightness has changed. The sent data packet contains information about the pixel coordinates, the polarity of the change (change to lighter or darker), and a time stamp. The camera operating in this way has a number of advantages over classical cameras that sequentially send pixels belonging to the entire image frame, e.g., high temporal resolution, low latency, no redundant data, and high dynamic range. The main goal of the proposed research is the acceleration of event-based dynamic vision sensor data processing. Attempts will be made to propose hardware architectures for the most important components of a modern vision system for autonomous robots.