CalibRBEV: Multi-Camera Calibration via Reversed Bird's-eye-view Representations for Autonomous Driving
Abstract: Camera calibration consists of determining the intrinsic and extrinsic parameters of an imaging system, which forms the fundamental basis for various computer vision tasks and applications, e.g., robotics and autonomous driving (AD). However, prevailing camera calibration models pose a time-consuming and labor-intensive off-board process particularly in mass production settings, while simultaneously lacking exploration of real-world autonomous driving scenarios. To this end, in this paper, inspired by recent advancements in bird's-eye-view (BEV) perception models, we proposes a novel automatic multi-camera Calibration method via Reversed BEV representations for autonomous driving, termed CalibRBEV. Specifically, the proposed CalibRBEV model primarily comprises two stages. Initially, we innovatively reverse the BEV perception pipeline, reconstructing bounding boxes through an attention auto-encoder module to fully extract the latent reversed BEV representations. Subsequently, the obtained representations from encoder are interacted with the surrounding multi-view image features for further refinement and calibration parameters prediction. Extensive experimental results on nuScenes and Waymo datasets validate the effectiveness of our proposed model.
Primary Subject Area: [Content] Media Interpretation
Relevance To Conference: Autonomous driving applications represent a crucial domain for multimedia information processing. Within autonomous driving systems, various sensors such as cameras, LiDAR, and radar are employed to perceive the surrounding environment of the vehicle, converting the perceived information into digital signals for processing. Among these, precise environmental perception (e.g. BEV perception) is of paramount importance for the safety and performance of autonomous driving systems. Camera calibration, therefore, stands as a pivotal task to ensure that the multi-view multimedia images captured by the cameras can be accurately interpreted and understood, thereby providing a reliable foundation for the system to make correct decisions and actions. (i) This work introduces the concept of Bird's Eye View (BEV) perception model into the calibration, achieving camera calibration tasks through a reversed BEV pipeline in autonomous driving, which presents an innovative and inspiring approach. (ii) This work proposes an autoencoder-based reconstruction method for bounding box data to achieve high-quality reversed BEV feature extraction, addressing the limitations of current forward BEV perception models in autonomous driving.
Submission Number: 1076
Loading