Automatic Detector of Gait Alterations using RGB-D sensor and supervised classifiers: a preliminary study

Alterations in walking patterns are widespread in the elderly population due to the motor decline typical of aging and other comorbidities related to movement disorders, such as Parkinson’s disease, or the consequences of acute events such as stroke. Early detection allows promptly activating specific rehabilitation treatments to reduce the risk of falls, injuries, and hospitalizations. This paper presents a non-invasive solution based on Azure Kinect and machine learning to detect gait alterations (i.e., slow-speed gait, short-step gait, and dangling gait). The body tracking algorithm captures the 3D skeletal model during gait on a straight walking path compatible with domestic environments. Some parameters are estimated from the virtual skeleton to characterize gait objectively. These parameters are then fed to supervised classifiers to distinguish between normal and altered gait (binary classification) and between types of alterations (multi-classes classification). Preliminary results obtained on healthy volunteers simulating alterations are presented and discussed.