Enhancing Reinforcement Learning in 3D Environments through Semantic Segmentation: A Case Study in ViZDoom