MORE | Fall 2022

Vehicle Speed Estimation from 3D Object Detection for Surveillance Cameras

One of the key components of an intelligent transportation system is the ability to detect and track vehicles from traffic surveillance cameras. This research project showcases a monocular 3D object detection system that is trained on a custom dataset, KITTI-Carla, which comprises synthetic 3D data collected from the view of a traffic camera from the Carla Simulation in the format of the well-known KITTI dataset. This method is more robust than 2D object detection because of the perspective transformation involved, thus yielding better speed estimation. The system will be tested on the BronoCompSpeed Dataset and deployed on the chipset of ARGOS vision’s camera stack which will calculate the speed of vehicles at traffic intersections in real time. The results would compare the accuracy and computational efficiency of the method proposed against already existing methods.