HMM-Fuzzy Model for Recognition of Gait Changes due to Trip-related Falls

This paper reports the use of HMM-based fuzzy rules generation for identifying the differences in gait between people with tendencies to fall and healthy people. This work is built on the work reported earlier by the authors where fuzzy rules were successfully applied in gait pattern recognition. This paper reports the hybridization of HMM with fuzzy logic for improving the recognition accuracy. Gait features were extracted from minimum foot clearance (MFC) data that was collected during continuous walking on a treadmill from 20 elderly subjects, 10 healthy and 10 with reported balance problem and history of falls. The input feature space was divided into a number of groups based on HMM generated log-likelihood values, and consequently each group was applied to construct a new fuzzy rule. Gradient descent method was used to optimize the parameters of the generated rules. These were then applied to recognize differences in the gait in subjects with trip-related falls history. The model's performance was evaluated using a cross- validation protocol applied on the training and testing data. The HMM-Fuzzy model outperformed the Fuzzy-based gait recognition as reflected both in the receiver operating characteristics (ROC) results as well as absolute percentage accuracy