Rozpoznávání činností řidiče z obrazů získaných vnitřní kamerou ve vozidle

Abstract

This master's thesis focuses on the recognition of driver activities using image data captured by an~in-car camera. The goal was to develop a system capable of automatically detecting actions such as fastening and unfastening the seat belt, shifting gears, and using a mobile phone. Driver posture was analyzed by detecting key body points using a skeleton-based approach with the MediaPipe library. The proposed features were then analyzed using recurrent neural networks of the Long Short-Term Memory (LSTM) type. The experiments showed that recognition of selected activities is feasible, while the achieved accuracy indicates room for further improvement.