Adaptive gait-event detection is essential for intelligent and autonomous control of walking assistive devices. Inertial measurement units (IMUs) based algorithms have been widely adopted but mostly suffered from technical limitations, such as time-delays and/or calculation burden, diverse gait parameters involving different speeds, and other factors in real-world applications. Those can lead to the mismatching of gait between the wearer and the device, further affecting wearing comfort and safety. To reduce the delay and improve the robustness of gait-event detection for knee assistive devices, a lightweight adaptive gait detection method based on an onboard machine learning approach is developed and preliminarily evaluated in this work. The method consists of a two-level algorithm working independently of each other. The first-level algorithm outputs the current detection results at the frequency of 100 Hz, and the second-level algorithm trains and updates the parameters of the gait model for detection every 10 seconds. A total of six events were detected in real-time on a portable Raspberry Pi with two IMUs on thigh and foot, including two specific knee-related events for eliminating detection delay with heel strike and toe off phases in walking. The proposed algorithm exhibits consistently high-performance scores (F1-score of events ≥ 0.92) and early detection capability (≤ 39 ± 22ms) at different walking speeds, in particular, the event prediction for heel strike and toe off were about 77 ± 10ms and 141 ± 10ms in advance, respectively. Given the simple and convenient hardware requirements, this method is especially suitable for intelligent knee assistive device applications.
In this work, a lightweight adaptive hybrid gait detection method with two inertial measurement units (IMUs) on the foot and thigh was developed and preliminarily evaluated. An adaptive detection algorithm is used to eliminate the pre-training phase and to modify parameters according to the changes within a walking trial using an adaptive two-level architecture. The present algorithm has a two-layer structure: a real-time detection algorithm for detecting the current gait pattern and events at 100 Hz., and a short-time online training layer for updating the parameters of gait models for each gait pattern. Three typical walking patterns, including level-ground walking (LGW), stair ascent (SA), and stair descent (SD), and four events/sub-phases of each pattern, can be detected on a portable Raspberry-Pi platform with two IMUs on the thigh and foot in real-time. A preliminary algorithm test was implemented with healthy subjects in common indoor corridors and stairs. The results showed that the on-board model training and event decoding processes took 20 ms and 1 ms, respectively. Motion detection accuracy was 97.8% for LGW, 95.6% for SA, and 97.1% for SD. F1-scores for event detection were over 0.86, and the maximum time delay was steadily below 51 ± 32.4 ms. Some of the events in gait models of SA and SD seemed to be correlated with knee extension and flexion. Given the simple and convenient hardware requirements, this method is suitable for knee assistive device applications.