An On-Node Processing Approach for Anomaly Detection in Gait

A novel method is proposed for capturing deviation in gait using a wearable accelerometer. Previous research has outlined the importance of gait analysis to assess frailty and fall risk in elderly patients. Several solutions, based on wearable sensors, have been proposed to assist geriatricians in mobility assessment tests, such as the Timed Up-and-Go test. However, these methods can be applied only to supervised scenarios and do not allow continuous and unobtrusive monitoring of gait. The method we propose is designed to achieve continuous monitoring of gait in a completely unsupervised fashion, requiring the use of a single waist-mounted accelerometer. The user’s gait patterns are automatically learned using specific acceleration-based features, while anomaly detection is used to capture subtle changes in the way the user walks. All the required processing can be executed in real-time on the wearable device. The method was evaluated with 30 volunteers, who simulated a knee flexion impairment. On average, our method obtained ∼ 84% accuracy in the recognition of abnormal gait segments lasting ∼ 5 s. Prompt detection of gait anomalies could enable early intervention and prevent falls.