Optimal Shoulder Joint Position Estimation Using Extended Kalman Filter based SLAM

JIITA, vol.6 no.2, p.574-578, 2022, DOI:

Bokku Kang1), Jong Cheol Lee2), Jae Seong Yoon2), Hyunki Lee2,*)

1) School of Mechanical Engineering, College of Engineering, Kyungpook National University, Daegu, South Korea
2) Division of Intelligent Robotics, Daegu Gyeongbuk Institute of Science & Technology, Daegu, South Korea

Abstract: Skeleton tracking algorithms[1] can be applied in various fields such as rehabilitation, motion capture, motion recognition, and sports. Recently, skeleton tracking algorithms based on RGB-D cameras such as Microsoft Azure Kinect and Intel RealSense cameras have been commercialized. In addition, it is applied in various fields such as motion capture, sports, and exercise, etc. However, the limitations of RGB-D camera-based skeleton tracking algorithms are also clear. Even if other drawbacks are excluded, still it is a clear limitation that it is difficult to make accurate measurements. Nevertheless, RGB-D-based skeleton tracking algorithms are considered to have great market potential thanks to their universal appeal that anyone can use them easily. This paper proposes a new optimization system using sensor fusion of RGB-D cameras and optical tracking systems to compensate for the shortcomings of RGB-D camera-based skeleton tracking algorithms. RGB-D type skeleton tracking algorithms used Nuitrack SDK[2], and optical tracking system (Skadi, Digitrack Inc., Daegu, Korea)[3] is used as marker based tracking system. The Extended Kalman Filter SLAM algorithm[4] was applied as a new optimization system, and segment length was used as mapping data considering that segment length was a value that did not change. A micro Lie group theory was applied to calculate the jacobian[5] considering the linear velocity and angular velocity of segment on 3D coordinate. This result was verified using a multi-joint human body mannequin model. As a result of the simulation, the accuracy of this algorithm for segment length prediction was improved and the accuracy of location prediction was also improved compared to when using Nuitrack SDK, a commercial skeleton tracking algorithm. In particular, due to the feature of the RGB-D camera-based skeleton tracking algorithm, the problem of outliers has also improved.

Keywords: Skeleton tracking; Sensor Fusion; Extended Kalman Filter; SLAM.

Scroll to top