Medical Small-scale Transportation Object Detection Algorithm Based on Multimodal Data Fusion

The emergence of the Covid-19 pandemic has greatly impact transportation, and unmanned transportation has been widely used in medical. The average precision of object detection as an important part in unmanned medical transportation. Object detection mainly relies on sensors of vehicles to obtain information about the surrounding obstacles like camera and LIDAR. In this paper, we introduce a new fusion way to fuse data from different modalities, as 2D and 3D object detection encouraging performance, they are typically based on a single modality and are unable to leverage information from other modalities. We leverage the geometric semantic consistency of 2D and 3D detection to obtain more accurate fusion results, and address the weaknesses of IoU in fusion network by using a generalized version as both a new loss and a new metric. The experimental evaluation on the challenging KITTI object detection benchmark, shows significant improvements in average precision, especially at bird’s eye view metrics, which shows the feasibility and applicability of the network.