Abstract: This paper designs and implements an elevator door status monitoring system based on multimodal perception and artificial intelligence analysis. The system integrates high-resolution cameras, microphones, laser rangefinders, pressure sensors, and ultrasonic sensors to achieve 24/7 non-contact real-time data acquisition and status analysis during normal elevator operation. The system was deployed in two buildings with significantly different structural types and usage intensities, collecting and analyzing data from over 6000 elevator operating cycles. Using a YOLO object detection model, the system identifies anomalies in landing doors and car doors through image analysis. Combined with brightness thresholding, root mean square （RMS） analysis of audio signals, and K-means clustering, the system effectively detects key fault modes such as door misalignment, slow door movement, structural noise, and floor localization errors. Experimental results demonstrate that the system exhibits strong adaptability and stability, providing reliable data support and a technical pathway for intelligent elevator monitoring and predictive maintenance.