Benchmarking Flex, FSR, and IMU Sensors for Wearable Posture Monitoring Using Deep Learning

Chronic back pain and musculoskeletal diseases are significantly exacerbated by poor posture, especially in sedentary people. Although wearable posture monitoring systems provide a useful option for ongoing evaluation and early correction, current systems use a variety of sensing modalities without a common framework for determining which sensor is best for precise and long-term posture measurement. Three widely used sensors for wearable posture assessment, Flex sensors, Force Sensitive Resistors (FSRs), and Inertial Measurement Units (IMUs), are compared in this study. To allow for simultaneous data collection, every sensor was combined into a single wearable belt and positioned in the same way. The gold-standard reference for measuring joint angles was a goniometer. During controlled activities like sitting, and forward bending, a real-time dataset was gathered from 250 people, including those with normal posture and those with problems linked to posture. Four deep learning models, Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), Transformer, and BiLSTM with sensor fusion, were used to simulate the nonlinear and temporal features of posture inputs. By comparing sensor outputs with goniometer readings, performance was assessed using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and temporal stability. IMUs are the most dependable option for wearable posture management belts because the results show that IMU-based measurements consistently achieve greater accuracy, robustness, and stability across models and posture scenarios.