Abstarct: Automatic license plate recognition problem from different aspects of the early 90s were studied. There are still many challenges for license plate recognition, such as fast moving vehicles at the scene, angles and different distances, so complex and unpredictable backgrounds, poor quality images and rotation of them, scaling, existence of multiple plates within an image, non-uniform and variable lighting throughout the day, and so on. In this thesis, a real-time algorithm is designed and implemented for detection and recognition of plates in video sequences, and simultaneous recognition of multiple plates in a video frxame. Already, in the field of detection and recognition of one plate in a scene is being done works that in most of them has been little attention to systems with real-time algorithms, while acceleration problem of recognition with correct detection and recognition of multiple plates on the scene is very important for its applications. Unlike methods with high computational complexity, we apply simple and effective techniques for being real-time. First, frxames containing moving objects are obtained by using Gaussian Mixture Model (GMM). Then, candidate regions are found by vertical edge detection and horizontal projection. After that, license plates are localized and extracted by morphological operations and connected components analysis. When plates were detected, their characters are separated with another algorithm and recognized plate numbers with license plate recognition system. Finally, the plates are recognized. The proposed method is evaluated on videos from highways cameras and correct detection rate is obtained 98.79 %. This system is implemented in C++ using OpenCV library. The average processing time of 25 ms per frxame and the overall average processing time 40 ms per frxame was achieved that can be used in real-time applications. Correct recognition rate is obtained 97.83 %. Our real-time system can also recognize multiple plates of different types in each frxame. Experimental results show that our method and how to implement have higher speed and better recognition rate than previous works therefore it is suitable for real-time applications.