YOdar: Uncertainty-based Sensor Fusion for Vehicle Detection with Camera and Radar Sensors

In this work, we present an uncertainty-based method for sensor fusion with\ncamera and radar data. The outputs of two neural networks, one processing\ncamera and the other one radar data, are combined in an uncertainty aware\nmanner. To this end, we gather the outputs and corresponding meta information\nfor both networks. For each predicted object, the gathered information is\npost-processed by a gradient boosting method to produce a joint prediction of\nboth networks. In our experiments we combine the YOLOv3 object detection\nnetwork with a customized $1D$ radar segmentation network and evaluate our\nmethod on the nuScenes dataset. In particular we focus on night scenes, where\nthe capability of object detection networks based on camera data is potentially\nhandicapped. Our experiments show, that this approach of uncertainty aware\nfusion, which is also of very modular nature, significantly gains performance\ncompared to single sensor baselines and is in range of specifically tailored\ndeep learning based fusion approaches.\n