Opportunistic calibration of sensor orientation using the Kinect and inertial measurement unit sensor fusion

Abstract

Sensor misplacement is a common obstacle that prevents inertial-based technology from providing reliable motion inference. Traditional approaches require certain calibration postures or activities to be performed. However, this may not be feasible for patients with mobility impairments. We propose a system that uses the Kinect's measurement as the ground truth to opportunistically detect and compensate for such errors. The goal of this study is to provide reliable motion data without the requirement of calibration activities or careful placement of the wearable sensors. First, we identified the instances where the Kinect had an unobstructed view of the limb of interest, and collected data for calibration. Then, we applied double exponential smoothing on the Kinect's position data and performed differentiation twice to generate virtual accelerations. By examining the acceleration vectors from the Kinect and inertial measurement unit (IMU) sensor, the misplacement of IMU sensors can be identified and thus compensated. Our results showed that the calibration algorithms successfully detected orientation error and provided accurate compensation. We also present an example of trajectory reconstruction with misplaced sensors and applied the proposed method. We obtained good agreement of reconstructed trajectories between the rectified sensor and the correctly placed sensor. The outcomes of this research will simplify ground-truth collection in the clinic, and provide reliable inference of motion data in the community.