Object detection has been widely adopted across various applications, but traditional methods mainly provide isolated object locations without capturing their relationships. To address this limitation, we propose a system that detects both objects and their relationships within images based on natural language queries. The approach integrates OWL-ViT for open-vocabulary object detection, RelTR for relationship inference, and Large Language Models (LLMs) for query processing and language understanding. Unlike fixed-vocabulary models, OWL-ViT enables detection from free-text descriptions, improving generalization and supporting flexible user queries. Experimental results show that the proposed framework can localize objects and infer their relations with a recognition accuracy of 27%, demonstrating its potential for intelligent systems such as query-driven surveillance and human–machine interaction. This work is not merely a combination of existing models, but a deliberate integration designed to address the novel challenge of detecting object relationships from natural language queries.