Transparent objects have been broadly applied in optical lenses, liquid crystal displays, airplane windshields, etc. During production and transportation, various defects may appear on the surfaces of transparent components, such as pits, scratches, and scuffs. In general, surface defects change the surface topography and transform the optical characteristics, thereby seriously degrading the structural stability and functional optimality of transparent components. Consequently, surface defect detection is essential for the quality evaluation of transparent-component-based systems. However, in principle, the parasitic reflection caused by the refracted light reflected by the back surface will inevitably induce extra detection errors or even false detection. To eliminate the parasitic reflection, we propose a single-view surface defect detection method with digital fringe projection, and an optical detection system is built to prove the feasibility of the proposed method. Comparative experimental results verify that the proposed method can accurately classify and locate surface defects of transparent components under the interference of parasitic reflection.
In our work, we propose a single-view surface imaging method based on reflection to eliminate parasitic reflection with digital fringe projection. A fringe-projection-based optical model is first built under the effect of fringe modulation by analyzing the difference between incident luminous flux and captured luminous flux. The model indicates that the topography variation of double surfaces can be reflected by two topography parameters, which constitute the average and contrast amplitudes of a cosine distribution. Subsequently, a multi-frequency temporal iterative technique is presented to precisely fit the cosine distribution during practical single-view surface imaging. In the practical iterative process, multiple fringe patterns with different frequencies are projected onto a transparent object. Then, the cosine distribution composed of intensity and frequency values is precisely fitted. Moreover, an iterative initial value acquisition algorithm is developed to achieve convergence in a short time. Finally, the average and contrast amplitudes can be applied to extract the topography parameters, thereby further identifying whole-field surface defects of transparent components and eliminating parasitic reflection.
The comparative experimental results of surface defect detection for the front surface (Figs. 6-8) verify that the proposed method avoids the ghost image and the low average intensity, thereby performing better. Since our method separates the topography information of different surfaces, it can effectively eliminate interference from the back surface. Only the topography information of the front surface remains, and thereby the ghost image is removed. Moreover, the contrast between defect regions and entire areas is increased under the proposed method, which improves detection performance. The comparative experimental results of parasitic reflection separation (Figs. 9 and 10) verify that the proposed method can reliably separate the topography information of front and back surfaces and avoid false detection. Besides, the detection results of the proposed method can show the defects clearly, and the contrast between defect regions and entire areas is increased under the proposed method, improving detection performance. Overall, the advantages of the proposed method are summarized as follows. 1) Simple detection system setup. The detection system is established based on the traditional reflection-guided detection method: only a camera, a light source, and a black background are needed. In addition, the proposed method is independent of assumptions on the surface shape or positional relations between equipment. 2) High-precision defect detection. In light of geometrical optics, photometry theory, and the optical characteristics of defects, a hybrid reflective-refractive model is built to analyze the incident luminous flux affected by double-sided transparent objects. Moreover, a multi-frequency iterative separation algorithm is designed to separate the topography information of different surfaces. A specific initial value acquisition algorithm is developed to acquire precise convergence results speedily. 3) In-situ detection of double surfaces. In our method, the equipment is fixed during detection, and only a single viewpoint is needed, making the proposed method suitable for in-situ defect detection. Experimentally, our method successfully utilizes the extra topography information contained in the mutual interference of double surfaces and eliminates parasitic reflection.
In this paper, a novel single-view surface defect detection method for transparent objects is proposed that can eliminate parasitic reflection with digital fringe projection. A mathematical model of hybrid light paths for transparent objects is built in light of geometrical optics, photometry theory, and the optical characteristics of defects. This model reveals that the topography parameters of double surfaces together constitute the average and contrast amplitudes of a cosine distribution. Moreover, a temporal iterative technique is presented to fit the cosine distribution. In the practical iterative process, multi-frequency fringe patterns are projected onto a transparent object. Then, the cosine distribution composed of intensity and frequency values can be precisely fitted. Finally, the topography parameters can be extracted from the average and contrast amplitudes, thereby further identifying whole-field surface defects of transparent components and eliminating parasitic reflection. To prove the feasibility of the proposed method, we build an optical detection system and conduct comparative experiments between the proposed method and other advanced methods. Experimental results verify that the proposed method can accurately classify and locate surface defects of transparent components under the interference of parasitic reflection.
