In the realm of Intelligent Transportation Systems (ITS), the integration of Computer Vision and Artificial Intelligence is pivotal. Even though the Highway Capacity Manual (HCM) recommends a range of speeds with respect to Free Flow Speed (FFS), in many nations, the emphasis is primarily on maximum speed limits. This traditional dependence on fixed/static speed limit systems, which frequently leads to congestion and inefficiency, fails to capture the dynamic nature of urban traffic. Our study presents a novel way to improve traffic control tactics by fusing sophisticated analysis with computer vision. To address traffic issues caused by the lack of variable speed limits on multi-lane urban roadways, we have used k-means clustering to analyze traffic flows at various points during different time periods. The clustering process can identify distinct patterns. We have analyzed these patterns using a Markov model-based system to predict future traffic conditions. The analysis involves essential features: Level of Service (LOS) and Percentage Speed Reduction (PSR). To validate the proposed method, we have collected over 25 h of video recordings from three different locations in the city of Bhubaneswar. These recordings have been collected over two time intervals, namely the morning (9:00 AM to 11:00 AM) and afternoon (1:00 PM to 2:00 PM). We have also studied how different traffic patterns affect the behaviour of various types of vehicles. This offers some insights on how to take targeted measures to improve traffic flow and enhance road safety. The code supporting this research is available at https://github.com/arabindaiitbbs/urban-traffic-prediction-clustering-markov .