Lokalizácia volantu vo vozidle pomocou kamery

Abstract

This bachelor thesis focuses on the design and implementation of a system for steering wheel localization in a vehicle using camera-based image data. The main objective is to detect the steering wheel and analyze the presence of the driver's hands on the steering wheel under real-world conditions. Deep learning methods were employed, specifically YOLO-based models for object detection and segmentation. Hand detection was performed using the MediaPipe Hands library, which enables localization of hand keypoints. As part of the work, a custom dataset of vehicle interior images was created and annotated. Experimental results showed that the detection model achieved an accuracy of approximately 97.07~\%, while the segmentation model reached approximately 98.08~\%. The proposed system for hand state classification achieved an accuracy of approximately 82.84~\% with a processing speed of about 4.57 FPS. The results indicate that the proposed solution is suitable for use in advanced driver assistance systems.