Human-like Observation-inspired Universal Image Acquisition System for Complex Surfaces in Industrial Product Inspection


연구 분야: Infrastructure



학회: Machine Intelligence Research


초록

Industrial surface inspection is crucial for the manufacture of high-end equipment across industries, with precise image acquisition being fundamental. Existing imaging systems often lack flexibility, they are restricted to specific objects, and face challenges in industrial structures without standardized computer-aided design (CAD) models or with complex surfaces. Inspired by human-like multidimensional observation, this study developed a universal image acquisition system based on measured point clouds, offering strong adaptability and robustness in complex industrial settings. The system is divided into three layers: the physical layer responsible for hardware integration, the interaction layer that facilitates bidirectional data exchange with the control layer, and the control layer integrating a new paradigm of multiple intelligent algorithms. The physical layer incorporates 2D and 3D cameras, turntables and industrial robots, enhancing the flexibility and compatibility of imaging. The interaction layer manages bidirectional information transmission and data exchange, offering a visualized area to enhance the user interaction experience. The control layer consists of point cloud preprocessing, primitive segmentation, viewpoint generation and pose estimation algorithms, using point cloud-based viewpoint generation and trajectory planning for high-precision image acquisition applicable to complex surface inspections across scenarios and structures. The system’s utility is demonstrated through a software and hardware algorithm platform and an interactive interface. Experimental validation on curved surfaces of different configurations and sizes confirms its universal image acquisition advantages. This system promises to introduce a cost-effective, versatile solution for complex surfaces, driving adoption across diverse industrial scenarios.


Author Profile
Tianbo Yang

Institute of Automation Chinese Academy of Sciences Beijing 100190 China

China
Author Profile
Shaohu Wang

The School of Artificial Intelligence University of Chinese Academy of Sciences Beijing 100049 China

China
Author Profile
Yuchuang Tong

Engineering Laboratory for Intelligent Industrial Vision Chinese Academy of Sciences Beijing 100190 China

China

📄 논문 정보

발행 연도 2025년
인용수 0
출판 국가 China
사이트 Springer
좋아요 수 0

연관 논문 목록 (239건)