Vehicle-3D-Lite: A Network for 3D Localization of Vehicles with Edge Lighting

The use of monocular 3D vehicle localization is vital in intelligent transportation systems (ITS) and cooperative vehicle coordination infrastructure systems (CVIS), encompassing areas like autonomous driving, analyzing traffic flow, and monitoring vehicle behavior. Nonetheless, due to the inability of monocular cameras to directly record depth data, conventional techniques often depend on intricate geometric limitations to ascertain the vehicle's 3D location, leading to not just heightened computational intricacy but also diminished processing effectiveness. In this paper, we propose a novel lightweight 3D positioning network for automobiles, which uses MobileNetV4 as the backbone network, while localization accuracy positioning accuracy and greatly reducing the model complexity and the consumption of computing resources. Experimentally, the model performed well in the vehicle-3D-lite localization task with an average accuracy of 68.44 % and real-time processing speed of 128 frames per second, 54 frames higher than Baseline. By reducing resource demand and improving processing efficiency, it provides a practical and broad application prospect for intelligent transportation infrastructure, especially edge computing scenarios.