Cradle Care: A Multi-Modal Infant Monitor using Audio-Video Fusion for Cry Detection and Alerting

Infant distress monitoring is a critical requirement for ensuring child safety and timely caregiver intervention. Traditional monitoring systems rely primarily on audio-based detection, which often results in unreliable performance due to ambient noise and lack of contextual understanding. This paper proposes a multi-modal deep learning framework that integrates both audio and video streams for robust infant cry recognition in real time. The audio pipeline utilizes Mel- Frequency Cepstral Coefficients (MFCCs) and a Convolutional Neural Network (CNN) to classify cry sounds, while the visual pipeline employs a YOLOv11- based object detection model to detect facial crying expressions. A decision fusion mechanism incorporating temporal averaging and confidence thresholds is implemented to minimize false alarms and ensure consistent detection before triggering an alert. Furthermore, an automated notification system is integrated using the Mailgun API to deliver real-time alerts to caregivers. Experimental evaluation demonstrates that the proposed multi-modal system achieves superior accuracy, reliability, and responsiveness compared to single- modality approaches, while maintaining computational efficiency suitable for real-world deployment in homes and neonatal care environments.