Recognizing the characters from a vehicle’s license plate is important for automobile tracking, traffic management, and parking area data management. This research develops a framework to detect and recognize Bangladeshi number plates automatically from CCTV cameras. The YOLO V7 model and several image processing techniques are used to detect the number plate in its number plate detection state. Then, localizing characters is performed using image processing techniques such as adaptive thresholding, dilation, connected component analysis, and so on. Several pre-trained CNN models have been trained and evaluated for the recognition phase. Then, we compare the results of these models to get an optimal technique for our framework. YOLO V7 performed excellently in the plate detection phase and had an accuracy of 0.91, precision of 0.994, recall of 0.901, and map at 0.5 of 0.938. Character localization is calculated manually and has 0.85 precision and 0.97 recall rate. In the final phase, from several deep learning models, VGG19 outperforms and has 0.936 accuracy, 0.912 precision, and 0.947 recall rate.