Intelligent traffic management has a considerable significance due to the rapid increase of the number of vehicles on our roads. Large number of vehicles cause many car accidents and traffic jams. These issues can be alleviated by using vehicle counting systems. In this thesis, two methods for vehicle counting are suggested. The first one uses deep learning algorithms. The YOLOv3 detection system is used to detect vehicles on the road. The YOLOv3 shows state-of-the-art results in the area of object detection. Vehicle trajectories were obtained using the ORB algorithm. ORB stands for Oriented F AST and Rotated BRIEF and is used to extract features from an image. Finally, trajectory analysis is performed to determine the driving detection and to count vehicles. Deep learning algorithms are widely used nowadays because they can solve many problems. However, they are very complicated and computationally demanding, which is unlikely to perform well on restricted hardware. The second method used in this thesis uses traditional computer vision methods. Using these methods showed to be more than ten times faster than using deep learning algorithms. Vehicle detection is performed using the Gaussian mixture model. Vehicle tracking is done using the Kalman filter. This algorithm, also known as linear quadratic estimation, produces estimates of unknown linear variables. Using this property, it can be used to track vehicles. Finally, the trajectory analysis is performed, just as in the first method. At the end of this thesis, the results of each method are shown and the comparison between the two methods is given