Enhancing Postural Monitoring in Wheelchair Users Through Context Classification

Globally, the number of wheelchair users is steadily increasing. These people often adopt sitting patterns that reflect their functional status. Monitoring the user's postural status can help users and healthcare professionals to treat them. However, this posture is sometimes influenced by the environment in which the chairs move, and not necessarily by changes in their functional status. To address this problem, this study presents a model designed to classify wheelchair movement contexts, enabling the identification of what is happening in the user's environment. To do this, data has been collected using a robust and non-intrusive combined monitoring system, which records both the wheelchair's movement and the user's posture. These data have been used to train classifier models capable of distinguishing between seven categories of environments that are common in the daily lives of wheelchair users: flat surface, ramp up, ramp down, right turn, left turn, obstacles, and abrupt braking. These models have been developed using machine learning techniques, such as K-Nearest Neighbors (KNN), Artificial Neural Networks (ANN) and Support Vector Machines (SVM). The results show an accuracy of 90% in free-running tests and more than 99% in controlled runs. These results remained consistent despite variations in training subjects, validated by leave 2 out cross-validation. This innovative approach has the potential to improve the quality of life of wheelchair users by providing an accurate and effective tool to understand and address complex interactions between the environment and the users' posture.