Deep learning-driven head posture detection and real-time feedback system

A real-time head posture detection and feedback system was developed to address increasing cervical spine problems in office environments. The system employs a lightweight HPENet model (4.2M parameters) achieving 3.35° detection error and 23.1FPS processing speed on Jetson Nano. Incorporating visual, auditory, and data recording feedback mechanisms, it improved user posture efficiency by 47.3%. The optimized system operates at 5.7W with 4.5-hour continuous runtime. A 72-hour stability test confirmed reliable performance across various environments, offering a practical solution for protecting computer users’ posture during extended work periods.