We introduce a geometric method for generating Euler angles from 3D landmark key-points predicted by Mediapipe, Openpose, or Tensorflow.js, with the aim of reconstructing 3D human posture. Our approach enables the use of predicted joint key-points from monocular motion capture applications to drive virtual humanoid avatar. While Mediapipe can detect 2D and 3D joint key-points of the entire body from a single image, it lacks rotation parameters for virtual character joints. To overcome this, we establish local coordinate systems for each joint using the results obtained by Mediapipe. As the joint rotates, the dominant bone’s coordinate axis also rotates. We initially compare the initial coordinate systems of each joint in the virtual character and then calculate rotation angles for each joint along the x, y, and z axes. These angles are combined in the specified order to determine the Euler angle. In this paper, we discussed the conversion method between the upper arm and hand, achieving high accuracy; Due to the concise formula, the algorithm complexity is very low. This method demonstrates the feasibility of using geometric algorithms to convert predicted results into full body inverse kinematics.