In this paper, we present a comprehensive methodology to support IoT based semantic traffic monitoring in smart cities to derive the status of traffic on a road segment in order to optimize traffic flow during rush hours. Smart traffic monitoring involves real-time traffic scene analysis from CCTV cameras to detect and identify stalled vehicles and measure traffic density on roads. The scene analysis essentially necessitates tracking and counting the number of moving and stopped vehicles as they progress along the image sequence by utilizing computer vision techniques. The obtained features along with manually annotated traffic data are translated to a higher level information using semantic web technologies, thus providing a common platform to exchange the information. As the perceptual modelling of traffic data involves media patterns so we have employed Multimedia Web Ontology Language (MOWL) for the semantic interpretation and to tackle inherent uncertainties involved in these media documents. MOWL supports hierarchical probabilistic reasoning framework and utilizes time-varying Dynamic Bayesian Networks (DBN) to perform semantic comparison and predict the evolution of changing situations. In this paper, we discuss the key tasks of vision and probabilistic reasoning components that provide a feasible solution to identify the cause of traffic jam. The proposed approach can be used to trigger an automatic warning or alerts to traffic authorities in case any anomalous activity is encountered and guide users to take some different route to avoid jam. The experimental results shows effectiveness of real-time vehicle monitoring to assess congestion on road and offer user an assistive environment to operate. © 2017 IEEE.