Human pose detection in 2D/3D images plays a vital role in a large number of applications such as gesture recognition, video surveillance and human robot interaction. Joint human pose estimation in the 2D motion video sequence and 3D facial pose estimation is the challenging issue in computer vision due to noise, large deformation, illumination and complex background. Traditional directed and undirected graphical models such as the Bayesian Markov model, conditional random field have limitations with arbitrary pose estimation in 2D/3D images using the joint probabilistic model. To overcome these issues, we introduce an ensemble chaining graph model to estimate arbitrary human poses in 2D video sequences and facial expression evaluation in 3D images. This system has three main hybrid algorithms, namely 2D/3D human pose pre-processing algorithm, ensemble graph chaining segmented model on 2D/3D video sequence pose estimation and 3D ensemble facial expression detection algorithm. The experimental results on public benchmarks 2D/3D datasets show that our model is more efficient in solving arbitrary human pose estimation problem. Also, this model has the high true positive rate, low false detection rate compared to traditional joint human pose detection models.