Low-cost autonomous car level 2: Design and implementation for conventional vehicles

Abstract

Modern cars are equipped with autonomous systems to assist the driver and improve driving experience. Driving assist system (DAS) is one of the most significant components of a self-driving vehicle (SDV), used to overcome non-autonomous driving challenges. However, most conventional cars are not equipped with DAS, and high-cost systems are required to equip these vehicles with DAS. Moreover, the design of DAS is very complex outside of the industry while it requires going through the Electronic Control Unit (ECU), which has a high level of security. Therefore, a basic system needs be installed in conventional cars which makes driving more efficient in terms of driver assistance. In this paper, an intelligent DAS is presented for real-time prediction of steering angle using deep learning (DL) and raw dataset collected from a real environment. Furthermore, an object detection model is deployed to assist and warn the driver of various types of objects along with corresponding distance measurement based on DL. Outputs from DL models are fed into the steering control system, which has Electronic Power Steering (EPS). The steering angle is measured in real time using an angle sensor and is posted back to the steering control system to make automated adjustments accordingly. Real-time tests are conducted on a 2009 Toyota Corolla equipped with a digital camera to capture live video stream, Controller Area Network (CAN-BUS) messages, and a steering angle sensor. The performance evaluation of the proposed system indicates intelligent steering control and driver assistance when evaluated in a real-time environment.

Keywords

• Self-driving vehicle
• Deep learning
• Image processing