SmartECushion: Design of a solution to detect remote working postures based on sensors and Deep Learning models

The increasing use of computers at home - driven by telecommuting, virtual education, and digital entertainment - has led to a growing incidence of musculoskeletal disorders (MSDs). These conditions often begin with pain in the neck, shoulders, back, and lower back, and are associated with prolonged adoption of improper postures without rest or active breaks. Such sedentary habits not only compromise physical well-being but also reduce productivity and contribute to increased healthcare costs and absenteeism. SmartECushion is a comprehensive solution for detecting improper postures during remote work. It integrates an ergonomic cushion embedded with ten force sensors, an Arduino Mega 2560 + ESP32 module for data acquisition, and a hybrid deep learning pipeline combining MLP and CNN architectures. Implemented using TensorFlow 2 and Keras 3, the model features a CNN branch to capture the temporal dynamics of sensor data and an MLP branch that incorporates user demographic information (gender, height, weight, BMI, and waist circumference). The model achieves 87.7% accuracy and a weighted F1-score of 87.5% on data from unseen users. For deployment, ONNX inference is served via a FastAPI microservice, decoupled from an automatic retraining service, all orchestrated using Docker containers to ensure scalability and continuous model updates.