Monocular 3D Object Detection using YOL011 and MiDAS for Autonomous Driving

Monocular 3D object detection is a critical component in autonomous driving systems, aiming to infer spatial properties from single-camera inputs. This paper presents a method that integrates YOL011 for 2D object detection and MiDaS for depth estimation to construct accurate 3D bounding boxes. By employing a calibration process and leveraging depth-aware tilting mechanisms, the model achieves precise object localization in real-world settings. The proposed approach demonstrated competitive accuracy on the KITTI dataset, with significant improvements in computational efficiency, making it suitable for real-time applications. Limitations in occlusion handling and depth extent estimation are discussed alongside future enhancement directions.