Real-Time Human Motion Capture Based on BiRNN and Inertial Sensor

This paper proposes a deep learning method using six sparse IMU inertial measurement units to realize body motion capture. It identifies the pattern of body posture task into multiple sub-tasks, estimates the whole body joint position from the blade joint position, and finally returns the rotation information of the whole joint to realize human posture recognition. For global translation tasks, RNN network fusion method are adopted to achieve better global translation effect. The method adopted in this paper compares the traditional visual method, and is not limited by space and funding. It can be captured through sparse IMU elements, and can achieve better results.