3D Human posture sequence estimation from single or multiple image sequences is essential in many applications, such as vision-based sport coaching and physical rehabilitation. However, 3D posture sequence cannot be accurately estimated from single image sequence due to depth ambiguity and self-occlusion, and pre-calibration is often required when estimating 3D posture sequence from multiple image sequences. In this paper, we present an algorithm to accurately estimate 3D human posture sequence from two un-calibrated image sequences by combining a modified Nonparametric Belief Propagation (mNBP) method with an efficient camera self-calibration method. The mNBP can estimate posture even when there is partial self-occlusion and when the human model scale is different from that of body image in image sequences. The efficient self-calibration can guarantee to find the optimal rotation and relative scale between two fixed but un-calibrated scaled orthographic cameras, without a nonlinear optimization process. Quantitative and qualitative test results show the algorithm's capability of estimating 3D posture sequence from a pair of un-calibrated image sequences.