This study presents a practical and efficient method for real-time speed estimation of vehicles on two-lane roads, particularly suited for surveillance applications. The proposed approach utilizes video input captured from a fixed, stationary camera without requiring prior camera calibration—making it both cost-effective and easy to deploy. The algorithm operates in two primary phases. In the initial phase, an interactive setup allows the user to manually define lane boundaries and corresponding real-world distances once at the beginning of the analysis. Based on this input, two rectangular regions of interest (ROIs) are assigned to each lane. In the second phase, moving vehicles are detected by generating an approximate binary foreground mask through frame differencing of consecutive video frames. By calculating the centroids of moving objects and measuring the norm values of the binary mask within each ROI, the time taken by each vehicle to traverse the space between the predefined lines can be determined. From this time and known distance, the average speed of each vehicle is estimated. Despite its algorithmic simplicity, the method achieves real-time performance without requiring specialized hardware. Experimental results demonstrate a mean speed estimation error of ±3 km/h and a vehicle detection accuracy of approximately 83%.