ANFIS Post-Processing for Real Time Gait Detection and Classification

Gait detection and distinction from other movement patterns like descending the stairs is a crucial task for an exoskeleton supporting user movement. Active or quasi-passive exoskeletons should enhance wearer's limbs only in a manner of not interfering with natural gait patterns. Common solutions for this problem are numerous gait detection algorithms that among other sensors use force sensing resistors. In this paper, we propose using an adaptive neuro-fuzzy inference system (ANFIS) classifier that can be trained on a stationary computer and only evaluated in a real time microprocessor control system. What is more, we propose altering the ANFIS outcome with five post-processing algorithms. Each network and algorithm combination is evaluated, results are compared and the best combined classifier is chosen.