Simultaneous Bayesian recognition of locomotion and gait phases with wearable sensors

Recognition of movement is a crucial process to assist humans in activities of daily living, such as walking. In this work, a high-level method for the simultaneous recognition of locomotion and gait phases using wearable sensors is presented. A Bayesian formulation is employed to iteratively accumulate evidence to reduce uncertainty, and to improve the recognition accuracy. This process uses a sequential analysis method to autonomously make decisions, whenever the recognition system perceives that there is enough evidence accumulated. We use data from three wearable sensors, attached to the thigh, shank, and foot of healthy humans. Level-ground walking, ramp ascent and descent activities are used for data collection and recognition. In addition, an approach for segmentation of the gait cycle for recognition of stance and swing phases is presented. Validation results show that the simultaneous Bayesian recognition method is capable to recognize walking activities and gait phases with mean accuracies of 99.87% and 99.20%. This process requires a mean of 25 and 13 sensor samples to make a decision for locomotion mode and gait phases, respectively. The recognition process is analyzed using different levels of confidence to show that our method is highly accurate, fast, and adaptable to specific requirements of accuracy and speed. Overall, the simultaneous Bayesian recognition method demonstrates its benefits for recognition using wearable sensors, which can be employed to provide reliable assistance to humans in their walking activities.