Abstract: Traditional manual supervision in drainage engineering construction management often suffers from insufficient oversight and overlooked safety risks. To enhance management efficacy, this study develops an intelligent management platform integrating IoT and computer vision technologies, enabling comprehensive site perception, data acquisition, intelligent analysis, and scientific decision-making. Trained and tested on a large - scale image library, an improved YOLOv5-based algorithm for personal protective equipment detection addresses safety gear identification challenges in complex construction environments and diverse worker postures. Besides, precise early-stage fire hazard recognitions are achieved by an improved flame and smoke detection Mask R-CNN algorithm. Experimental results demonstrate over 85% mean average precision in target recognition, effectively enabling behavior monitoring and risk identification. The convergence of IoT and computer vision technologies empowers the platform with real-time data synchronization and dynamic supervision capabilities, providing data-driven decision support that significantly improves resource allocation efficiency and safety management performance. This technological integration establishes a novel paradigm for intelligent infrastructure management, demonstrating practical value in modernizing construction supervision systems.