| Abstract [eng] |
The precise detection of objects and the surrounding environment is one of the essential requirements for applying computer vision systems in the automotive, robotic, and aviation industry. The quality of such systems has a strong influence on the actions of autonomous vehicles on the road. To avoid traffic accidents, autonomous vehicles must accurately identify other road users and obstacles. Due to the wide variety of objects and environmental conditions encountered on the road, image segmentation remains one of the most challenging tasks in the field of computer vision. This thesis examines computer vision models for image segmentation, their architectures, and the data sets used, with the aim of creating an efficient system to segment road scene imagery. Based on an analysis of the most effective existing models, an image segmentation algorithm built upon a transformer-based neural network architecture is implemented. In the experimental section, modifications applied to the system are described and compared, and their impact on the accuracy of the image segmentation system is assessed. |
