HumanVLM: Foundation for Human-Scene Vision-Language Model