Can a Rollator Predict Your Posture? The Potential of Pose Estimation and Depth Image Analysis For Posture Analysis

A smart rollator walker with integrated sensors offers a promising solution for improving the mobility and safety of older people. By recording and analyzing the user’s posture, potential fall risks can be identified at an early stage. This study presents a method for posture classification while using a rollator walker by integrating semantic information from RGB images with geometrical data of depth maps. In the first step of the method, human pose detection is achieved using YOLO pose to identify key points, specifically the hip joints, which serve as reference landmarks for defining a region of interest (RoI) in the corresponding depth maps. Structural features within the RoI are extracted using gradient-based techniques to highlight areas with significant depth variations. These features, along with statistical measures like mean depth value and standard deviation form the final feature vectors for posture classification. Various machine learning models, including Support Vector Machines (SVM), Logistic Regression (LogReg), Random Forest (RF), and an ensemble of XGBoost and Random Forest (XGRF), are employed to differentiate between good and bad postures. The model trained on proposed features demonstrate strong performance, with the RF models and XGRF achieving the highest accuracy of 0.85, precision of 0.89 and 0.88, and recall of 0.83, indicating reliable and balanced classification of good and bad postures.