In this paper, we propose an early prediction method for work operations to estimate the destination of the hand from the initial motion of the worker. A feature of the work operation is that it consists of grasping actions of parts and tools placed around the worker. Therefore, the positions of these objects are considered to provide effective information for early prediction of actions. In this study, we propose a spatial graph, Human-Object Relational Graph (HORG), which describes the positional relationship between human joints and objects, and apply it to Graph Convolutional Networks. Experimental results using an original dataset consisting of 6,900 grasping actions (about 26k frames) by 23 subjects showed that the proposed method achieved a recognition rate of 66.8% at 0.36 seconds before the end of the motion, which is 8.5% higher than that of the conventional method (ST-GCN). The advantage of our method was also confirmed in terms of the positional accuracy of the hand position prediction.

In the visual inspection of industrial products using a microscope with a shallow depth of field, it is difficult to capture an all-in-focus image. This paper presents a method to generate an all-in-focus image from a large focal stack, which can be applied to visual inspection. The proposed method adds two improvements to a deep learning method [M. Maximov et al., CVPR, (2020) 1068] that leverages defocus cues in depth estimation. First, it interpolates an index map (a collection of in-focus image indices at all pixels) after the forward pass. It generates an all-in-focus image using all images in focal stacks while reducing the number of images input to the network. Second, the network is trained with synthetic datasets to which a random texture with high-frequency components is added. It helps the network to learn the degree of defocus blur. Experiments using synthetic images and real microscope images show that interpolating the index map and adding texture improves the accuracy of all-in-focus image generation.

MEMS scanners use micro torsion bars made of single crystal silicon. The push-in method is used to measure the mechanical properties of these micro torsion bars, in which a lever connected to the torsion bar is pushed. In the push-in method, the sliding force increases as the indentation amount increases, decreasing measurement accuracy. As a method to reduce the sliding force, we proposed to generate pure torsion in the torsion bar section by rotating a sample holder with a test sample attached. The proposed method can maintain the load detection axis of the load cell and the lever part at almost a right angle regardless of the degree of torsional angle, preventing slip at the contact point between the load cell and the lever part. The torsional spring constants of test samples were obtained and compared by three different methods: a testing system using this principle, a resonance frequency method, and a theoretical calculation. The accuracy of the developed testing system is -17.5 % to 12.7 % of the measurement error compared to the torsion beam spring constant obtained from the resonance frequency, and -15.6 % to 7.2 % compared to the theoretical calculation value, thus verifying the effectiveness of the proposed test method.