In this paper, we propose a novel, real-time unsupervised learning based event discovery framework for indoor environment. We assume that a static camera is continuously observing an indoor area. Our framework is able to find the important regions such as the entrances and exits which we call as hotspots and to find the most popular paths. Our framework also finds the correlations between these hotspots. We develop an incremental clustering mechanism to discover events on-the-fly and define a probability associated with each cluster to ensure the correctness of the clustering stage.