Evaluation of the Distance between the Self-Driving Vehicle and the Point of Impact Using the Carla Simulator

Abstract

The assessment of road safety often relies on measuring the distance required to detect objects as a metric for evaluating potential hazards. In freeway simulation models, determining the distance between two objects plays a crucial role in the decision-making process of autonomous vehicles. It helps evaluate the level of interaction between vehicles. However, calculating the necessary distance for object detection is a complex task. It requires predicting future vehicle interactions and planning optimal routes for the target vehicle and other associated vehicles to anticipate potential collisions. This study aims to explore and evaluate computation methods for incorporating this additional distance into object recognition within simulation-based models for microscale simulations. A specific approach is recommended, and the results of the conducted experiments are presented. The computations leverage vector-based kinematic factors and dependency calculations on bounding boxes. The computational complexity and execution time of the proposed method are assessed. The findings of these experiments demonstrate promising potential for comprehensive and effective investigations. It is concluded that a combined calculation approach is effective, despite its high cost, inconvenience, unreliability, and challenges in testing. To overcome the limitations of real-world vehicle research, we propose an efficient and cost-effective object detection technique based on the CARLA simulator. CARLA provides a biologically-inspired system of activities centered around the local area, which aids in the development, training, and validation of autonomous driving systems. Compared to other methods, our approach outperforms in terms of precision, recall, and f1 score. For instance, while yolov4 achieves 92.5%, 78.2%, and 84.7% respectively, our method achieves 96.6%, 76%, and 85%. This paper provides an overview of the CARLA simulation, which has been purposefully developed to support the creation, training, and certification of automated vehicles. CARLA offers free access to open digital assets specifically designed for this purpose, including urban environments, buildings, and vehicles. The simulator platform enables customization of sensor suites, the operating environment, full control over static and dynamic entities, map construction, and other functionalities.