The method of recognizing traffic signs through image processing has increased in popularity along with advanced driver assistance systems. Drivers may have difficulty reading and detecting traffic signs due to fatigue, weather conditions and speed while driving. In our study, traffic signs rectangular, square, circle and so on. Regardless of the type of different plates seen in the country, even if the correct detection is aimed. By sending the model as a parameter while training, the only thing that needs to be done within the scope of adding a new plate is to retrain our model. Before starting learning, the image was enhanced to improve the performance of the algorithm by using the Contrast Restricted Adaptive Histogram Equation (CLAHE) method in data processing. In our study, results were obtained with 2 deep learning models unlike classical CNN architecture. VGG-16 and Xception deep learning models were compared with each other. SGD and Adam optimization methods were tried for both models and the optimum method was found for our study. Our study has reached an accuracy value of up to 98.38%. The speed performance of our method is sufficient to enable a real-time system implementation in the future. In order to understand the results of our experimental tests in the system to be used, it has been turned into a return parameter and the driver can be integrated with the vehicle regardless of the screen and used with voice assistant or small structures to be added independently of the vehicle.