Evaluace metod pro lokalizaci těla řidiče ve vozidle

Abstract

This bachelor thesis focuses on a comparison of methods for locating driver body parts inside a vehicle. In the introductory part, the basic concepts and methods in the field of object recognition in images are described, with particular emphasis on approaches designed to detect human limbs and joints. Furthermore, available training and test datasets taken from real vehicle environments are examined; in the absence of a suitable database, the creation of a custom dataset is proposed. A program is then presented that uses at least three selected human body part localization methods (e.g. OpenPose, AlphaPose, YOLO or MediaPipe) and compares them in terms of functionality, accuracy and speed. The conclusions of the paper summarize the advantages, shortcomings and future development opportunities of these methods for use in autonomous driving systems that can apply driver location analysis in crisis situations.