• Civil engineering
• Coordinate measurement machine
• Experiment virtualisation
• Intersystem comparison
• Non-destructive testing
• Uncertainty quantification

• convolutional neural network
• deep learning
• fringe-to-phase transformation
• single-shot imaging
• three-dimensional image acquisition
• three-dimensional sensing

• convolutional neural network
• deep learning
• fringe-to-phase transformation
• single-shot imaging
• three-dimensional image acquisition
• three-dimensional sensing

• 3D digital models
• 3D ultrasonic tomography
• carbonate materials
• close-range photogrammetry
• petrographic analyses
• terrestrial laser scanner

• convolutional neural network
• defect detection
• photometric stereo

• Photogrammetry/methods
• Algorithms
• Photometry

• high-resolution 3D reconstruction
• industrial metrology
• non-collaborative objects
• photogrammetry
• photometric stereo
• point cloud

Product defect inspections are extremely important for industrial manufacturing processes. It is necessary to develop a special inspection system for each industrial product due to their complexity and diversity. Even though high-precision 3D cameras are usually used to acquire data to inspect 3D objects, it is hard to use them in real-time defect inspection systems due to their high price and long processing time. To address these problems, we propose a product inspection system that uses five 2D cameras to capture all inspection parts of the product and a deep learning-based 2D convolutional neural network (CNN) with spatial and channel attention (SCA) mechanisms to efficiently inspect 3D ball joint socket products. Channel attention (CA) in our model detects the most relevant feature maps while spatial attention (SA) finds the most important regions in the extracted feature map of the target. To build the final SCA feature vector, we concatenated the learned feature vectors of CA and SA because they complement each other. Thus, our proposed CNN with SCA provides high inspection accuracy as well as it having the potential to detect small defects of the product. Our proposed model achieved 98% classification accuracy in the experiments and proved its efficiency on product inspection in real-time.

• attention mechanism
• convolutional neural networks
• deep learning
• quality inspection

This paper presents an automatic system for the quality control of metallic components using a photometric stereo-based sensor and a customized semantic segmentation network. This system is designed based on interoperable modules, and allows capturing the knowledge of the operators to apply it later in automatic defect detection. A salient contribution is the compact representation of the surface information achieved by combining photometric stereo images into a RGB image that is fed to a convolutional segmentation network trained for surface defect detection. We demonstrate the advantage of this compact surface imaging representation over the use of each photometric imaging source of information in isolation. An empirical analysis of the performance of the segmentation network on imaging samples of materials with diverse surface reflectance properties is carried out, achieving Dice performance index values above 0.83 in all cases. The results support the potential of photometric stereo in conjunction with our semantic segmentation network.

• deep learning
• image processing
• photometric stereo
• quality control
• semantic segmentation

The acquisition of the geometry of general scenes is related to the interplay of surface geometry, material properties and illumination characteristics. Surface texture and non-Lambertian reflectance properties degrade the reconstruction results by structured light technique. Existing structured light techniques focus on different coding strategy and light sources to improve reconstruction accuracy. The hybrid system consisting of a structured light technique and photometric stereo combines the depth value with normal information to refine the reconstruction results. In this paper, we propose a novel hybrid system consisting of stripe-based structured light and photometric stereo. The effect of surface texture and non-Lambertian reflection on stripe detection is first concluded. Contrary to existing fusion strategy, we propose an improved method for stripe detection to reduce the above factor’s effects on accuracy. The reconstruction problem for general scene comes down to using reflectance properties to improve the accuracy of stripe detection. Several objects, including checkerboard, metal-flat plane and free-form objects with complex reflectance properties, were reconstructed to validate our proposed method, which illustrates the effectiveness on improving the reconstruction accuracy of complex objects. The three-step phase-shifting algorithm was implemented and the reconstruction results were given and also compared with ours. In addition, our proposed framework provides a new feasible scheme for solving the ongoing problem of the reconstruction of complex objects with variant reflectance. The problem can be solved by subtracting the non-Lambertian components from the original grey values of stripe to improve the accuracy of stripe detection. In the future, based on stripe structured light technique, more general reflection models can be used to model different types of reflection properties of complex objects.

• blinn-phong model
• photometric stereo
• reflectance property
• stripe detection
• stripe structured light

We present photometric stereo algorithms robust to non-Lambertian reflection, which are based on a convolutional neural network in which surface normals of objects with complex geometry and surface reflectance are estimated from a given set of an arbitrary number of images. These images are taken from the same viewpoint under different directional illumination conditions. The proposed method focuses on surface normal estimation, where multi-scale feature aggregation is proposed to obtain a more accurate surface normal, and max pooling is adopted to obtain an intermediate order-agnostic representation in the photometric stereo scenario. The proposed multi-scale feature aggregation scheme using feature concatenation is easily incorporated into existing photometric stereo network architectures. Our experiments were performed with a DiLiGent photometric stereo benchmark dataset consisting of ten real objects, and they demonstrated that the accuracies of our calibrated and uncalibrated photometric stereo approaches were improved over those of baseline methods. In particular, our experiments also demonstrated that our uncalibrated photometric stereo outperformed the state-of-the-art method. Our work is the first to consider the multi-scale feature aggregation in photometric stereo, and we showed that our proposed multi-scale fusion scheme estimated the surface normal accurately and was beneficial to improving performance.

• computer vision
• convolutional neural network
• deep learning
• photometric stereo

• Adenomas
• Colonic polyps
• Colonoscopy
• Image enhancement
• Machine vision
• Photometric stereo