Real-Time 3D Position Estimation from Monocular Cameras to Enhance Sensor Data Sharing for Extended Recognition Range in Autonomous Vehicles

Sharing sensor data from infrastructure cameras to autonomous vehicles in V2X-based C-ITS can enhance the safety of autonomous driving. However, 2D object recognition using a single camera is prone to distortion according to camera angles. To mitigate distortion, 3D position estimation becomes crucial. While many existing methods generate depth maps for images and perform 3D object recognition, real-time data sharing, a requirement for autonomous driving, remains challenging. In this paper, we propose a method that efficiently estimates 3D positions using a 3D position estimation model trained on data annotated with 3D information for objects recognized through 2D object recognition. Experimental results show the effectiveness of the proposed method.