(I) Objective: Develop a pose estimation system to identify baby key points, assisting parents in monitoring their baby's activities and well-being.

(II) Approach: Selected the YOLOv7-Tiny Pose model to meet real-time frame-per-second (FPS) requirements on low-compute-power chips used in the company’s products.

(III) Dataset: Curated a diverse dataset of baby images from various sources, including Google, YouTube, and client-provided data, to ensure the model effectively learns various baby poses.

(IV) Results:

• The YOLOv7-Tiny Pose model successfully identified baby key points in RGB images during the day and near-infrared (NIR) images at night.

• Demonstrated reliable performance under diverse lighting conditions while maintaining real-time processing capability.