Cloud computing technology is integral to the advancement of intelligent transportation systems. Integrating cloud computing into intelligent transportation systems needs to proceed with caution however, as cloud computing introduces new layers of security risks. As intelligent transportation systems generally rely on captured images of private citizens, security of these images is paramount. In this thesis, we propose an efficient system for improving the security of a cloud-based intelligent transportation system built with road side units, and a data collection and analysis server. A convolutional neural network is used to classify encrypted images obtained by roadside units based on the type of vehicle on the road in real-time, leaving personal information in these images hidden throughout the process. The proposed system never fully decrypts the collected images, thus protecting drivers’ personal information, such as location, license plate, and vehicle contents. As the system never needs to fully decrypt the images, the system increases efficiency compared to a system which fully decrypts the images for analysis. The results show improved computational performance in comparison with a fully decrypting system, while keeping the data secure.