Phase measuring deflectometry (PMD) is widely used to measure the surface of specular objects. The planar screen used in traditional PMD systems have the drawbacks of a small field of view of measurement and a large virtual image depth range. This problem can be effectively solved by using the curved screen. As the arc structure of the curved screen is different to the planar structure required in Zhang’s method, the existing methods mainly rely on the stereo vision of two cameras to obtain the screen position. This paper presents a novel PMD based on a curved screen and single camera. A method of calibrating a curved screen by constructing a virtual planar screen is presented. The method only needs one camera to calibrate the curved screen directly with high accuracy. In addition, the change from planar screen to curved screen makes the traditional phase-gradient-height relation invalid. Based on the comprehensive analysis of space optical path models, a novel phase-gradient-height relation is presented. The relation is a general formula. It can be used for any curved screen, including the planar screen. Simulation and quantitative comparison experiments verify the accuracy and reliability of the proposed calibration method. The measurement results of concave mirror show that the proposed method has higher accuracy and a larger field of view in measuring specular objects than traditional methods.

Whereas standard classifiers fail to achieve substantial results in dealing with under-represented instances. In this paper, a method of improving SMOTE, BDSMOTE(Fusion of boundary factors and data density for improved SMOTE) is proposed. In this method, the concepts of data density and boundary factor are introduced. Firstly, minority samples are divided into boundary samples and safety samples based on data density, and each minority sample is given initial weight, the safety sample region is combined with more samples, and the boundary sample region is relatively less. Secondly, the boundary factor is introduced to update the sample weight, so that the number of synthetic instances of each sample is different. Then compose a new instance with the SMOTE method. Finally, six pretreatment methods are compared on random forest classifier. Experiments on KEEL and UCI data sets show that the classification model using the proposed method and random forest has better classification performance than the other five classification models.

The major problem with 3D room layout reconstruction is estimating the 3D room layout from a single panoramic image. In practice, the boundaries between indoor objects are difficult to define, for example, the boundary position of a sofa and a table, and the boundary position of a picture frame and a wall. We propose MreNet, a novel neural network architecture for predicting 3D room layout, which outperforms previous state-of-the-art approaches. It can efficiently model the overall layout of indoor rooms through a global receptive field and sparse attention mechanism, while prior works tended to use CNNs to gradually increase the receptive field. Furthermore, the proposed feature connection mechanism can solve the problem of the gradient disappearing in the process of training, and feature maps of different granularity can be obtained in different layers. Experiments on both cuboid-shaped and general Manhattan layouts show that the proposed work outperforms recent algorithms in prediction accuracy.