Automatic human activity detection is one of the difficult tasks in image segmentation application due to variations in size, type, shape and location of objects. In the traditional probabilistic graphical segmentation models, intra and inter region segments may affect the overall segmentation accuracy. Also, both directed and undirected graphical models such as Markov model, conditional random field have limitations towards the human activity prediction and heterogeneous relationships. In this paper, we have studied and proposed a natural solution for automatic human activity segmentation using the enhanced probabilistic chain graphical model. This system has three main phases, namely activity pre-processing, iterative threshold based image enhancement and chain graph segmentation algorithm. Experimental results show that proposed system efficiently detects the human activities at different levels of the action datasets.