Real-Time Onboard Human Motion Recognition Based On Inertial Measurement Units

Locomotion motion recognition is important in gait analysis and control of wearable robots to achieve smooth gait transitions. In this paper, we propose a support vector machine based locomotion intent prediction system using two Inertial Measurement Units (IMUs). The prediction system can classify locomotion modes in daily life onboard online. Two IMUs were put on the right front of thigh and shank of the subject respectively, and each of them generated three channels of angles, three channels of accelerations and three channels of angular velocities. To evaluate the performance of the system, several experiments are conducted on three able-bodied subjects for five activities including sit, sit-to-stand, stand, level-ground walking, and stand-to-sit. Average recognition accuracy is 94.25% +/- 0.72%. Most transitions can be detected before hand and no missed detections are observed for all the trials.