Accurate detection of buildings in urban remote sensing imagery is essential for urban planning and management. However, challenges persist due to the dense distribution of buildings and the limited boundary localization precision of existing methods. To address these issues, this study proposes a rotated bounding box detection method based on intersection-over-union (IoU) supervision. The proposed IoU-based localization accuracy supervision network effectively suppresses interference from adjacent buildings in complex scenes and enhances the localization accuracy of detection boxes. In addition, a circular smooth label strategy is introduced to mitigate loss oscillations caused by the periodic nature of angular variables. Experimental results demonstrate that the proposed method achieves a mean average precision of 93.31% and a recall of 95.19% on the Building-OBB dataset, and an accuracy of 76.93% on the DOTA dataset, validating its effectiveness.