Abstarct: In this thesis, a real time algorithm for speed estimation of on-road vehicles is proposed. To determine the speed of vehicles, two video processing methods are proposed and evaluated using the OpenCV library and C ++ programming language. The first method is baxsed on extraction of background model and correlation algorithm. In this method, the mixture of Gaussian is used to obtain background model. Then we select an area of the image for speed measurement. In order to increase the accuracy and eliminate the nonlinear distortion of the image, a projective transformation is applied to deperspective this area. Presence of a vehicle is verified using the correlation between this area in the background model and the upcoming frxame. Finally, speed is estimated by dividing the length of this area by the presence time of the vehicle in it. In the second method, we develop a real-time algorithm, so we avoided complicated tracking algorithms. In this method, at first, the area of interest for speed estimation is determined. This area is rectified by the projective transformation. The initial model of the foreground, including movable objects, is obtained by the difference of successive frxames. By morphological operations, irrelevant objects are eliminated so that moving objects become more coherent. The distance traveled by each vehicle is measured by the displacement of the center of gravity in successive frxames within the specified area. Travel time is obtained by counting the number of frxames. Eventually, car speed is obtained by dividing the displacement by time for each car and multiplying the result by a factor that converts pixels per second to km/s. The both methods are implemented on Lenovo z510 laptop with a 2.5 GHz Intel Core i5 processor and 6 GB of RAM, as well as on an Odroid XU4 with 2 GHz 8-core processor and 2 GB of RAM. The runtime was measured in the first method was 55 ms on laptop and 105 ms on XU4 board. The reported speed error was 4.99% with a standard deviation of 4.83% and a difference of ± 2.26 km / h. In the second method, the runtime on the laptop was 31 ms and on the XU4 Board was 48 ms. Also, the reported speed error was 3.31% with a standard deviation of 3.28% and a difference of ± 1.39 km / h.