Detection and Analysis of Sitting Posture in Real Time Based on Keras Framework

Sustaining proper sitting posture is crucial in the contemporary era of sedentary lifestyles to prevent musculoskeletal problems and promote general well-being. Using the Keras framework, this study proposes a tailored method for the instant identification and evaluation of sitting positions. The suggested approach uses a pre-trained DenseNet201 model with frozen layers and custom layers on top to train a specialized model for identifying sitting postures. The method accurately categorizes sitting postures as good, bad, or undefined by using a wide dataset of sitting posture photos and transfer learning. The system’s ability to analyze and monitor seated postures in real time is improved by the integration of YOLOv3 for person detection and MediaPipe Holistic for pose estimation. The system provides immediate visual feedback, using color-coded indicators (green for good, red for bad, and blue for undefined postures), to assist users in self-assessing and correcting their sitting postures. Furthermore, the system incorporates notification alerts to prompt users when a bad posture persists, motivating them to make necessary adjustments. Experimental results demonstrate the effectiveness of the system in promoting healthy postural habits and reducing the risk of musculoskeletal issues associated with improper sitting posture. Future work involves exploring posture analysis techniques and expanding the system’s capabilities for a comprehensive analysis of sitting postures. This innovative approach addresses the growing concern of sedentary behavior by providing real-time posture monitoring and feedback, contributing to long-term postural health and well-being.