Stereo Vision Meta-Lens-Assisted Driving Vision

作者：Xiaoyuan Liu, Wuyang Li, Takeshi Yamaguchi, Zihan Geng, Takuo Tanaka, Din Ping Tsai, Mu Ku Chen · 发表于：ACS Photonics · 年份：2024 · DOI：10.1021/acsphotonics.3c01594 · 被引用次数：41 · 研究领域：CCD and CMOS Imaging Sensors、Advanced Optical Imaging Technologies、Image Enhancement Techniques

Object detection and depth perception are key foundations of object tracking and machine navigation, facilitating a thorough perception and understanding of the surrounding environment. Currently, autonomous vehicles employ complex and bulky systems with high cost and energy consumption to achieve demanding multimodal vision. An imperative exists for the development of compact and reliable technology to enhance the cost-effectiveness and efficiency of autonomous driving systems. Meta-lens, a novel flat optical device, has an artificial nanoantenna array to manipulate the light properties. It is lightweight, ultrathin, and easy to integrate, making it suitable for various applications. We developed a stereo vision meta-lens imaging system for assisted driving vision, a comprehensive perception including imaging, object detection, instance segmentation, and depth information. The compact system comprises a band-pass filter, a stereo vision meta-lens, and a complementary metal oxide semiconductor (CMOS) sensor. In comparison to traditional two-camera-based stereo vision systems, the meta-lens stereo vision imaging system eliminates the need for distortion correction or camera calibration. A tailored data processing pipeline is proposed with an intensity and depth gradient cross-validation optimization mechanism and three deep learning modules for object detection, instance segmentation, and stereo matching foundations. Final assisted driving vision provides multimodal perception...