Real Time Implementation of Inter-Car Distance Based on an Intelligent Stereovision System for Autonomous Vehicles

Mohamed Sejai, Anass Mansouri, Saad Dosse Bennani, Yassine Ruichek

Abstract: In recent years, the fusion of deep learning and computer vision technologies has significantly advanced the development of autonomous vehicles that are present more and more in road traffic. In this context, this paper proposes a vehicle vision system that combines two techniques, the first uses artificial intelligence algorithm to accurately identify vehicles in the path of vehicle’s trajectory, the second uses stereovision algorithm to precisely estimate inter-vehicle distances. This solution effectively reduces overall processing time by exploiting the advantages of the You Only Look Once real-time vehicle detection and limiting the region of interest in image to the computation area of the disparity map for the stereovision. Detection and distance estimation of numerous vehicles consumes an important computation time; therefore a parallel data processing based on the Open Multi-Processing library is used to optimize data processing performance. The proposed solution is implemented on an embedded platform, the experiment results show that the system successfully detects vehicle and estimate distance with an error rate of less than 10%, achieving a real-time processing of 30 frames per second.