Sensor Fusion based Gait Recognition for Accurate Actuation of Foot Drop Stimulation

Abstract

Foot drop is a walking (gait) disability resulting with defect in neural pathway to lower limb after a stroke or traumatic event. The foot drop patients mainly lack the forefoot lift (dorsiflexion) while swinging the foot forward. In some cases, excessive inward movement (inversion) during swing and insufficient pressing against the ground (plantar flexion) before the swing is also accompanied. Due to these attributes, the patients require external support for a safe and efficient walk. Functional electrical stimulation (FES) is a prominent technique to assist with gait of such patients. In FES, the peroneal nerve and lower leg muscles are stimulated by electric current to promote required movement at ankle joint prior swing. A sensory mechanism is therefore employed that detects the gait and actuate stimulation at the right instance. However, the gait sensing system in commercial foot drop stimulators exhibit inappropriateness in stimulation triggering under certain communal scenarios. Beside triggering appropriateness, stimulation adaptation is another feature that is required in certain situations and commercial foot drop stimulators (FDSs) do not cater for that, as well. Existing works endeavoured to overcome both types of stimulation control limitations, however, situations like adaptation with changes in walking direction are still to be addressed. This thesis presents a gait recognition system that is based on unique bipedal gait model (BPM). The BPM is formed by cross product of finite state model (FSM) of gait at each foot. The sensing platform used for gait and activity information relies on force sensitive resistors (FSRs) for plantar foot pressure; accelerometers for foot acceleration and velocities; and gyroscopes for foot angular motion. A stimulation control algorithm is further developed that determines when to actuate stimulation upon the gait acknowledgement from the sensing setup. This sensing and control system was evaluated by real-time simulation under the situations causing inappropriate actuation in commercial FDS. The system proved successful for accurate stimulation control as well as identification of activities other than walking e.g. resting, standing. After BPM, the thesis introduces a real-time detection of walking direction, which is based on dominant trend duration (DTD) in anteroposterior acceleration. Next, the BPM driven stimulation control algorithm is updated with directional information. Thereafter, the augmented sensing and stimulation control is tested on a bidirectional walking data in a real-time simulation manner to evaluate adaptive control of stimulation triggering with respect to walking direction. The results affirm capability of the stimulation control for adaptation with back stepping alongside normal forward walk.