IMU-Based Gait Event Detection Using Classification Methods

Gait event detection has been one of the topics of interest in biomechanics in recent years. This research topic is useful in rehabilitation and biomedical engineering. In this study, different phases during walking will be recognized using a single IMU and also two FSR sensors to label data as a wearable device. First, we will calculate Separation Index (SI) to find out the possibility of dividing the gait cycle into four phases via kinematic data. Then, different classifiers would be learned to be able to separate phases. Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Fine Tree, and Adaptive Neuro-Fuzzy Inference System (ANFIS) would be employed for this purpose. After data processing, the SI which is 0.7655 proved the possibility of classification. Then, our experimental data will feed into classifiers. Results show that KNN method yields better performance (84.8 percent accuracy). In addition, SVM, Fine Tree and ANFIS accuracies are 82.4, 80.7 and 79.5 percent respectively.

Keywords:

Gait Phase Detection , Wearable Device , Learning Classifiers , Inertial Measurement Unit