th and ?y is the estimation. Table 2 shows the mean and standard deviation of the relative L2 pose error norms on the walking sequences. Our approach outperforms the other state of the art algorithms =-=[24, 3]-=- in estimation accuracy 3 . And the computational speed of our approach 4 is 15 times faster than [24, 3] thanks to the integral histograms used in APC descriptor extraction and fast inference of huma