• dataset creation
• human detection.
• human pose estimation

This thesis analyzes existing datasets and model architectures for human detection and pose estimation, identifying limitations and suggesting improvements for real-time applications. It reviews and evaluates state-of-the-art datasets and models, formulating guidelines to address common issues like pose diversity and occlusion handling. A proof-of-concept dataset was created to demonstrate these guidelines, showing enhanced annotation detail and scenario coverage. Testing the model's efficacy using metrics like Intersection of Unions (IOU) highlighted the effectiveness of these enhancements. The research offers insights into optimizing datasets and model configurations, contributing to advancements in computer vision and artificial intelligence.