Analýza bezpečnostních pásů ve vozidle pomocí obrazů

Abstract

Recently, with the growing interest in vehicle safety and in-vehicle monitoring, the importance of systems that can analyze driver behavior and seat-belt usage has increased. This bachelor thesis focuses on developing an experimental software system capable of detecting the status of a driver’s seat belt through image analysis.

In the theoretical part, selected approaches to solving image analysis tasks are presented, with an emphasis on object detection methods in images. Particular attention is given to convolutional neural network architectures and modern detection models, especially the YOLO family, which is used in the practical part of the thesis.

The practical part of the thesis describes the implementation of an experimental software solution that analyzes seat belts in vehicle images. This includes the preparation and use of appropriate datasets, the design and training of the detection model, and the experimental verification of its functionality, accuracy, and speed. The work utilizes widely used frameworks, including OpenCV, TensorFlow, and PyTorch.