Towards a Wearable Wheelchair Monitor: Classification of push style based on inertial sensors at multiple upper limb locations

Measuring manual wheelchair activity by using wearable sensors is on the rise for rehabilitation and monitoring purposes. Stroke pattern is an important descriptor of the wheelchair user's quality of movement. This paper evaluates the capability of inertial sensors located at different upper limb locations plus wheel, to classify two types of stroke pattern for manual wheelchairs: semicircle and arc. Data was collected using bespoke inertial sensors with a wheelchair fixed to a treadmill. Classification was done with a linear SVM algorithm, and classification performance was computed for each sensor location in the upper limb, and then in combination with wheel sensor. For single sensors, forearm location had the highest accuracy (96%) followed by hand (93%) and arm (90%). For combined sensor location with wheel, best accuracy came in combination with forearm. These results set the direction towards a wearable wheelchair monitor that can offer multiple on-body locations for increased usability.