Abstract
Autonomous vehicles have been considered as one of the most important trending topics in the domain of intelligent transportation systems. It is expected that most of the leading companies will launch their fully self-driving vehicles by the end of the next decade. Such a technology must assure a high safety level on road networks to reduce the number of accidents caused by human errors. All the built-in technologies must be integrated and complemented to achieve these goals. Automated object and obstacle detection is one of the main research tasks that must be undertaken. A number of vision-based learning techniques have been designed to improve the vehicle detection capabilities and reduce the shortcomings of other sensors such as LIDAR system that have shown poor results during severe weather conditions. In this paper, we overview and investigate the main learning models for video-based object detection specifically that can be applied with autonomous vehicle. We focus on a machine learning solution, namely, the Support Vector Machine (SVM) algorithm and two deep learning solutions: the "You Only Look Once" (YOLO) and the Single Shot Multibox Detector (SSD) methods. Afterwards, we provide an exhaustive comparison of the different approaches when applied to autonomous vehicles. Simulations results indicate that the SVM poorly performs and its speed can not assure real-time response, while the YOLO model and SSD can reach higher accuracy with a notable ability to detect objects in real-time when rapid driving decisions need to be made.