This paper focuses on real-time rotation estimation for model-based automated visual inspection. In the case of model-based inspection, spatial alignment is essential to distinguish visual defects from normal appearance variations. Defects are detected by comparing the inspected object with its spatially aligned ideal reference model. Rotation estimation is crucial for the inspection of rotationally symmetric objects where mechanical manipulation is unable to ensure the correct object rotation. We propose a novel method for in-plane rotation estimation. Rotation is estimated with an ensemble of nearest-neighbor estimators. Each estimator contains a spatially local representation of an object in a feature space for all rotation angles and is constructed with a semisupervised self-training approach from a set of unlabeled training images. An individual representation in a feature space is obtained by calculating the Histograms of Oriented Gradients (HOG) over a spatially local region. Each estimator votes separately for the estimated angle; all votes are weighted and accumulated. The final estimation is the angle with the most votes. The method was evaluated on several datasets of pharmaceutical tablets varying in size, shape, and color. The results show that the proposed method is superior in robustness with comparable speed and accuracy to previously proposed methods for rotation estimation of pharmaceutical tablets. Furthermore, all evaluations were performed with the same set of parameters, which implies that the method require minimal human intervention. Despite the evaluation focused on pharmaceutical tablets, we consider the method useful for any application that requires robust real-time in-plane rotation estimation.
COBISS.SI-ID: 10508116
We present a framework for automated visual inspection of pharmaceutical tablets in heavily cluttered dynamic environments. A two light camera system is proposed, which acquires two images with different lighting directions. With two images, pose of each tablet in the scene is first estimated. Second, regions that are shadowed or overlapped by neighboring tablets are detected. The final analysis is then performed on the remaining areas of the tablets. The proposed framework was evaluated on a set of real pharmaceutical tablets, imaged with the proposed system inside a rotating drum. The rotating drum simulated the movement of the tablets inside the tablet coating machine. Acquired images were analyzed with the proposed framework and the analysis results were compared with a gold standard, which was prepared manually. With the proposed framework we detected 81% of defective tablets while 5% of good tablets were erroneously classified. The results indicate that the proposed framework is a viable approach for the on-line and in-line analysis of partly diffuse objects.
COBISS.SI-ID: 11323476
In this paper, we propose a method for pose estimation of multiple (textureless) objects of the same type in heavily cluttered environments. The method as such could be used as a component for building more flexible automated visual inspection systems, removing the need for precise mechanical manipulation and enable inspection in settings previously thought unfeasible. The method consists of three phases. In the first phase, template matching is used to calculate similarity measure maps for different object poses. Template matching combines both edge and surface normal information to improve the pose estimation accuracy. Second, a large number of pose hypotheses are generated with a nonparametric clustering of the similarity measure maps and finally, the best hypotheses are iteratively selected. Method was evaluated and compared with the current state-of-the-art on two synthetic and one real-world datasets. The results show that the proposed system performs better than the current state-of-the art for pose estimation in industrial environments.
In this paper, we present a method for real-time pose estimation of rigid objects in heavily cluttered environments. At its core, the method relies on the template matching method proposed by Hinterstoisser et al., which is used to generate pose hypotheses. We improved the method by introducing a compensation for bias toward simple shapes and by changing the way modalities such as edges and surface normals are combined. Additionally, we incorporated surface normals obtained with photometric stereo that can produce a dense normal field at a very high level of detail. An iterative algorithm was employed to select the best pose hypotheses among the possible candidates provided by template matching. An evaluation of the pose estimation reliability and a comparison with the current state-of-the-art was performed on several synthetic and several real datasets. The results indicate that the proposed improvements to the similarity measure and the incorporation of surface normals obtained with photometric stereo significantly improve the pose estimation reliability.
We have developed a new method for estimation of the coating thickness from images of cluttered and overlapping tablets.