Object tracking in video sequence is a common task of computer vision. Problem of generic object tracking is following: There is in advance unknown object in a video scene and our task is to keep its position during sequence, respectively keep object`s trajectory while it is moving within a video scene. We can find practical applications of object tracking in security monitoring systems, military guidance systems, video surveillance systems, statistical measurements, and nowadays it’s also widespread in autonomous vehicles. Currently there are various approaches and tens of algorithms that are dealing with this problem, one of the main goals of this thesis was to compare and review these methods. Tracking algorithms namely have to face number of challenges. Problems might arise while the tracked object is occluded by other objects or by background scene, merging with other objects or background, deformation of the object, other visual changes or sudden unpredictable movement etc. Therefore we conducted a comparison of selected methods to find out how do they deal with these challenges. We briefly looked through principles which are used by tracking algorithms, then we proposed metrics, testing dataset, we implemented testing tool for metrics evaluation and consequently we performed our tests. During review and testing, we identified one big potential for object tracking. It`s hidden in neural networks, which can bring excellent results in very short time. Considering tests’ results, we chose the best method, which we subsequently used for practical applications on traffic videos. First of them was tracking traffic signs in a sequence recorded by camera placed at the car`s front window, the second was tracking vehicles at the crossroads. As our task evolved from single object tracking to multi object tracking, we had to modify the chosen method. However, the method did not work well at these tasks, so we tried to improve it using external detector. Although this way we did not improve the results of the original tracking method, we created our own tracking method, which is based on detection joining and it can bring satisfying results in selected tasks.