Radar-Image Fusion Transformer for Object Detection with Uncertainty Estimation

Multi-sensor fusion is currently the main way of autonomous driving perception. Among them, radar has been paid more and more attention by researchers due to its low cost and strong anti-interference. The fusion perception method of radar and image has gradually become a hot research field. Nevertheless, the high-noise data of radar brings the challenge of uncertain perception results. In order to solve these problems, this paper proposes a Transformer-based 2D object detection algorithm with the fusion of radar and images, which can effectively detect objects. And for the problem of uncertainty in perception results, a method for modeling regression and classification uncertainty is proposed. We validate the proposed method on the NuScenes dataset. Experimental results show that our method achieves higher detection performance compared to the baseline model and can estimate the uncertainty of the proposed algorithm in real traffic scenarios.