A bionic compound eye-based 3D object detection method for intelligent driving

Vision-based 3D object detection in intelligent driving presents significant challenges, primarily due to the absence of depth information in vision systems. To address this limitation, a bionic compound-eye 3D object detection method tailored for intelligent driving is proposed, integrating depth information from compound-eye imaging into driving scenarios. First, inspired by the distributed perception mechanism of insect compound eyes, we design a novel bionic compound-eye array camera architecture for intelligent driving. Next, a compound-eye dataset collection method based on the CARLA virtual simulation environment is developed, resulting in the creation of a compound-eye intelligent driving dataset in nuscenes format. Finally, a multi-view feature fusion module specifically for compound-eye imaging is devised, leading to the development of the bionic compound-eye 3D object detection network (BCE-3D). On the compound-eye intelligent driving dataset, BCE-3D achieves 64.4% in mAP and 58.5% in NDS. The experimental results demonstrate the effectiveness of the bionic compound-eye array camera architecture and BCE-3D in intelligent driving 3D object detection. For more details on the compound-eye dataset collection method, refer to: https://github.com/ DarkerSun6/compound_eye_carla_nuscenes.