With the in-depth development of underground space utilization, double-shield tunnel boring machines (TBM) are widely used in hard rock tunnel construction due to their high efficiency, safety, and low environmental impact. However, in complex tunnel environments, existing guidance systems suffer from critical limitations, including poor environmental adaptability, complex structures leading to high costs and maintenance difficulties, reliance on bulky targets, and difficulty in balancing real-time performance and measurement accuracy. These issues restrict TBM excavation efficiency and construction safety, thereby becoming a key bottleneck for further development. Therefore, this study proposes a double-shield TBM guidance technology and develops a low-cost, high-precision, adaptable, and user-friendly guidance system based on binocular vision. It is of great significance for improving double-shield TBM pose measurement accuracy and stability and ensuring the reliability of tunnel construction.

Guided by binocular vision principles, the measurement system was developed through two fixed-spaced cameras, and triangulation and pixel matching were used to convert 2-D images into 3-D spatial data. Binocular cameras served as the core measurement unit, simplifying hardware by eliminating the need for multiple sensors. Feature points were installed on the front shield, binocular cameras were installed at the front end of the support shield, and laser targets were installed at the back end of the support shield. The binocular cameras were rigidly connected to the laser targets. By measuring the pose of the laser target using a total station and combining it with a coordinate system conversion algorithm, the final solution for the front shield pose was completed. Five coordinate systems were defined (Fig.1), and stepwise conversion (front shield→ target→ camera→ laser target→ total station) was used to derive front shield head coordinates. The moving average filtering method was adopted to mitigate vibration-induced noise. Three experiments were designed to verify system performance, including static repeatability test, dynamic absolute accuracy test, and on-site test under actual construction vibrations.

Static repeatability experiments showed that the X, Y, Z coordinate ranges of the front shield were 0.000346 m, 0.000350 m, and 0.00043 m, respectively, with standard deviations less than 0.00007 m and a maximum error less than 1 mm (Table 1, Fig.3), which reflected the static stability and measurement accuracy of the system. Dynamic tests revealed that both horizontal and vertical errors were less than 15 mm with reasonable data variation (Table 2), overcoming the limitation of insufficient dynamic accuracy in traditional methods. On-site tests demonstrated that the absolute errors of horizontal and vertical deviations of the shield head were less than 20 mm under intense vibrations, with stable and normal data (Table 3, Fig.4), effectively resisting dust and vibration interference to achieve continuous guidance. Compared to conventional guidance systems, the technology enhanced environmental adaptability via stereo matching, reduced costs through simplified hardware, optimized construction procedures by avoiding bulky targets, and balanced real-time performance and measurement accuracy with filtering and matching algorithms.

The binocular vision-based double-shield TBM guidance technology proposed in this study addresses several limitations of existing guidance systems. The technology accurately measures front shield pose through camera-feature point cooperation, coordinate conversion, and filtering. The experimental results verify that the system meets construction requirements for accuracy and stability, providing new insights for the design and optimization of double-shield TBM guidance systems. Future research can focus on optimizing camera calibration to streamline processes and improve precision.