Algorithm 1: Algorithm G2O-Pose

Input. 2D poses, camera parameters. Output. 3D poses. Step 1. Preprocess: calculate 3D bone segment proportions. Suppose the depth of the body (randomly) and calculate the lengths of bone segments. Step 2. Initialize the G2O optimizer, and the initial 3D pose. While (the depth change of the body is less than T), repeat Step 3–Step 7. Step 3. Calculate 3D body orientation. Step 4. Correct reverse joints of legs. Step 5. Solve 3D coordinates of the legs by optimization with the previous 3D pose as initial solution. Step 6. Correct reverse joints of the spine and arms. Step 7. Solve 3D coordinates of the upper part of body by optimization.