Vignarca Daniele, Vignati Michele, Arrigoni Stefano, Sabbioni Edoardo
Department of Mechanical Engineering, Politecnico di Milano, 20156 Milan, Italy.
Sensors (Basel). 2023 Aug 12;23(16):7136. doi: 10.3390/s23167136.
In recent years, the research on object detection and tracking is becoming important for the development of advanced driving assistance systems (ADASs) and connected autonomous vehicles (CAVs) aiming to improve safety for all road users involved. Intersections, especially in urban scenarios, represent the portion of the road where the most relevant accidents take place; therefore, this work proposes an I2V warning system able to detect and track vehicles occupying the intersection and representing an obstacle for other incoming vehicles. This work presents a localization algorithm based on image detection and tracking by a single camera installed on a roadside unit (RSU). The vehicle position in the global reference frame is obtained thanks to a sequence of linear transformations utilizing intrinsic camera parameters, camera height, and pitch angle to obtain the vehicle's distance from the camera and, thus, its global latitude and longitude. The study brings an experimental analysis of both the localization accuracy, with an average error of 0.62 m, and detection reliability in terms of false positive (1.9%) and missed detection (3.6%) rates.
近年来，目标检测与跟踪研究对于旨在提高所有道路使用者安全性的先进驾驶辅助系统（ADAS）和联网自动驾驶车辆（CAV）的发展变得至关重要。十字路口，尤其是在城市场景中，是道路上发生最相关事故的部分；因此，这项工作提出了一种车路协同（I2V）预警系统，该系统能够检测和跟踪占据十字路口并对其他驶入车辆构成障碍的车辆。这项工作提出了一种基于安装在路边单元（RSU）上的单个摄像头进行图像检测和跟踪的定位算法。借助一系列线性变换，利用相机内部参数、相机高度和俯仰角来获取车辆与相机的距离，进而获得车辆在全局参考系中的位置，从而得到其全球纬度和经度。该研究对定位精度（平均误差为0.62米）以及误报率（1.9%）和漏检率（3.6%）方面的检测可靠性进行了实验分析。
