Algorithm 1 A proposed refinement of initial human pose estimates

Input: initial pose estimate with initial joint positions $\left\{j_{i,k}\right\}$ (Section 4.1) Output: refined pose estimate 1:Bone Lengths Initialization (Section 4.2): The first N frames are used to calculate initial bone lengths based on symmetry and temporal constraints. Bone length hypotheses that strongly deviate for these initial bone lengths are down-weighted. 2:Trust data detection (Section 4.3): For new frames, the detected bone lengths are compared to the initial ones. If they deviate significantly or the involved joints are not tracked according to the Kinect SDK, the bone is marked as not reliably tracked. 3:Prediction of joint positions (Section 4.3): For not reliably tracked bones, we predict respective joint positions that are used as initialization to our optimization framework. 4:Pose refinement based on symmetry and kinematic constraints (Section 4.4): From the resulting joint positions, we compute a refined pose estimate based on exploiting symmetry constraints and bone length constancy. The optimization is performed using an energy minimization framework.