4 August 2025 in Measurement Science and Technology
With the continuous advancement of underwater detection technology, laser vision methods offer significant advantages in underwater environments, including high measurement accuracy and robust performance. However, current methods for refractive transformation of underwater images typically use small-angle approximations, neglecting the varying errors in different directions caused by the shift of the optical axis on the imaging plane. Additionally, three-dimensional (3D) reconstruction systems require complex underwater calibration to function properly in submerged environments. To address the challenges associated with refractive distortion in underwater 3D reconstruction, this paper proposes the laser stereo vision (LSV) based underwater 3D reconstruction system that corrects optical axis displacement caused by refraction-induced imaging shifts. First, a multi-medium imaging perspective transformation model is developed to correct refractive distortion. By analyzing the refraction path, this model provides a perspective transformation to convert refracted images into undistorted ones. Second, an underwater LSV based 3D reconstruction method is introduced that requires calibration solely in air. By combining stereo vision constraints and corrected images, an accurate 3D model of the underwater scene can be generated. Finally, a LSV underwater 3D reconstruction system was developed and tested. Image correction and 3D reconstruction experiments were conducted underwater. The average image correction error of the model was 0.08 pixels, and the average error along the Z-axis was 0.25 mm, with a standard deviation of 0.28 mm. Experimental results show that the proposed system can generate high-precision 3D point cloud models in underwater environments.
13 February 2025 in IEEE Transactions on Instrumentation and Measurement
Currently, concrete structural damage detection primarily relies on industrial cameras to capture 2-D images for identifying cracks and voids. However, assessing the extent of damage using these images is challenging. The existing volume measurement methods face significant errors and poor robustness due to irregularities in concrete defect formation in real-world scenarios. To address these challenges, this article proposes a quantitative analysis method for concrete defect volumes based on an improved point cloud slicing technique. Initially, high-precision 3-D cameras capture local point cloud data of concrete. Using a fusion method that integrates local curvature and relative distance, the areas showing defects in the concrete point cloud were successfully extracted. Subsequently, these defect areas undergo slicing, and an enhanced Euclidean clustering method effectively separates the contour boundaries of each slice. Further refinement involves combining the ray parity method and area effectiveness principle to accurately differentiate containment relationships among multicontour regions and calculate precise areas for each slice. By summing the areas of all slices, the volume of concrete defects is accurately determined. This article validates the proposed method’s accuracy and robustness in volume calculations through comparative experiments on standard specimens with different features and performs quantitative analysis of concrete defect volumes in real-world scenarios. Experimental results demonstrate that the proposed method outperforms the existing approaches in volumetric quantification of concrete defect point clouds, with an average measurement error of 3.08%. This represents a significant improvement over current methods and exhibits high robustness, providing a promising solution for evaluating the extent of concrete defect damage.
2 July 2025 in IEEE Transactions on Instrumentation and Measurement
The weak geometric features and high repeatability of information in tunnel scenes are important factors that make point cloud fine registration very difficult. This makes it challenging for descriptors to distinguish information differences between individual key points. To solve the registration issues caused by the scarcity of key points and high feature similarity in tunnels, this paper introduces a tunnel appendage structure (TAS) descriptor and a registration framework that improve the registration capability of the descriptor. The TAS descriptor encompasses two types of information for key points: the first is the local structure and primary direction of the tunnel, which belong to local feature information; the second is the positional relationship between the key point and the central axis, which belongs to global feature information. The combination of local and global features enhances the descriptive capability of the TAS descriptor, enabling it to accurately match corresponding key points. The registration framework divides the six-parameter registration problem into two three-parameter problems, thereby reducing the descriptive capability requirement of the descriptor in registration and improving algorithm efficiency. Experiments have shown that the method proposed in this paper is superior to existing advanced methods in terms of registration accuracy and speed in tunnel registration.