Development of an Attention-Based Convolutional Neural Network-Long Short-Term Memory Model for Real-Time Ergonomic Analysis of Sitting Posture

Abstract

The digital era has increased the prevalence of musculoskeletal disorders caused by poor sitting posture, posing a significant global health and productivity challenge. This study introduces an attentionbased deep learning model as the analytical engine for a proposed virtual ergonomics monitor, Ergo-Guard. The primary objective is to develop a model that accurately performs real-time Movement Quality Assessment of Sitting Posture for computer users, using only a standard webcam to ensure wide accessibility. This research method is a hybrid architecture that combines a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM), enhanced with an attention mechanism and optimized for three-dimensional skeletal data using the BlazePose Computer Vision approach. This framework merges a One-Dimensional CNN to extract spatial features from static poses with a Bidirectional LSTM network to model temporal postural shifts. An integrated attention mechanism enables the model to dynamically focus on critical ergonomic areas, mimicking an expert’s assessment. For validation, a new OfficePosture dataset was created, containing 500 videos of five common office sitting postures. The results indicate that the proposed model achieves 94.2% classification accuracy,
substantially outperforming baselines from a pure CNN (84.6%) and a standard LSTM network (89.2%). Beyond accuracy, the model offers interpretable feedback through visual attention maps. In conclusion, the proposed architecture is an effective solution for monitoring sitting posture and holds considerable promise as an affordable preventive health tool for corporate and educational settings.