Online Gait Phase Detection in Complex Environment Based on Distance and Multi-Sensors Information Fusion Using Inertial Measurement Units

The wearable powered exoskeletons have obtained great interest in enhancing the ability of people to perform heavy physical labor (e.g., soldiers transporting military supplies in high altitude areas and couriers carrying express goods in the logistics industry). A key problem we care about is how to make the exoskeletons provide efficient aids for people in complex environment (e.g., different terrains, different speeds, and different payloads), and the exoskeletons controller system needs the gait phase information reflecting people's walking state for creating aids for people. There are two main methods of gait phase detection, which are model-based and distance-based methods. Model-based approaches describe data distribution by employing discriminative (e.g., support vector machines (SVMs), recurrent neural networks (RNNs) and convolutional neural networks (CNNs)) and generative (e.g., hidden Markov models (HMMs) and Gaussian models) models. Distance-based approaches match locomotion pattern with reference sets by employing distance measures, such as dynamic time warping (DTW). The distance-based method during training is better when the available data is limited, because the model-based method will face the problem of over-fitting. This study proposes an online gait detection method based on distance and multi-sensors information fusion to solve the gait phase detection problem in complex environment. In addition, considering the force sensitive resistors (FSRs) limited lifespan, we adopt lower extremity angles measured by the IMU sensors as the input signal of the method for improving the performance. The results show that the method can adaptively detect the gait phase in real time for different terrains and different payloads while walking at different speeds. Our proposed method can gain over 95% accuracy and time difference is 4.99 +/- 15.05 ms.