CalibRBEV: Multi-Camera Calibration via ReversedBird's-eye-view Representations for Autonomous Driving.
Wenlong Liao, Sunyuan Qiang, Xianfei Li, Xiaolei Chen, Haoyu Wang, Yanyan Liang, Junchi Yan, Tao He, Pai Peng
Unverified — Be the first to reproduce this paper.
ReproduceAbstract
Camera calibration consists of determining the intrinsic and extrinsic parameters of an imaging system, which forms the fundamentalbasis for various computer vision tasks and applications, e.g., robotics and autonomous driving (AD). However, prevailing cameracalibration models pose a time-consuming and labor-intensive offboard process particularly in mass production settings, while simultaneously lacking exploration of real-world autonomous drivingscenarios. To this end, in this paper, inspired by recent advancements in bird’s-eye-view (BEV) perception models, we proposes anovel automatic multi-camera Calibration method via ReversedBEV representations for autonomous driving, termed CalibRBEV.Specifically, the proposed CalibRBEV model primarily comprisestwo stages. Initially, we innovatively reverse the BEV perceptionpipeline, reconstructing bounding boxes through an attention autoencoder module to fully extract the latent reversed BEV representations. Subsequently, the obtained representations from encoderare interacted with the surrounding multi-view image features forfurther refinement and calibration parameters prediction. Extensiveexperimental results on nuScenes and Waymo datasets validate theeffectiveness of our proposed model.