The main objective is to develop a roadside perception box capable of creating a real-world digital-twin map from roadside camera video streams. The solution must be modular, scalable, and real-time. To achieve this, a real-time object detection model must be selected. RF-DETR achieves state-of-the-art accuracy and latency trade-offs surpassing other top real-time models like YOLO11 and YOLO26. After detection, real-world (3D) coordinates must be estimated from pixel (2D) coordinates, which is challenging because most methods rely on stereo cameras, LiDAR, or radar systems. However, we have a single 2D frame that we need to map to 3D. A common method used in traffic speed cameras is the homography technique, which needs at least 4 real-world plane coordinates with their corresponding pixel coordinates.