Showing papers by "Michihiko Minoh published in 2016"

Intention-Sensing Recipe Guidance via User Accessing to Objects

Abstract: Sensing the intention of a user’s forthcoming action is a necessary function for systems that assist human physical activity. In this article, a strategy for recipe guidance systems that can predict the forthcoming intended subtask in a cooking task is investigated. The focus is on user accessing objects, that is, touching and releasing objects. Touching can indicate the start of the forthcoming subtask and releasing can indicate the end of the task. The main difficulty lies in the fact that humans may move objects because they are in the way and use cooking tools that are unanticipated by an assistive system. In such cases, the accessed object should not indicate the forthcoming subtask. A method is proposed to track the progress of a task based on the object access history. This enables to eliminate object accesses that are out of context. Simultaneously, the method predicts the forthcoming subtask based on a combination of progress and materials rather than tools and materials. Then, a guidance...

Kusk Object Dataset: Recording access to objects in food preparation

Abstract: This study aims to construct a KUSK Dataset's extension that provides records of chef's touching and releasing action to objects, which we call “access to objects,” in his/her food preparation. The records of access to object are known as a key evidence for understanding chef's activity in food preparation. In the dataset, we provide object images as well as the records of access to object. The data are obtained by manual annotation and by automatic processing. As a result of annotation, we collected 4391 object images from 57 cooking observations. We also confirmed that the CNN-based state-of-the-art method reached 74.15% accuracy on average in recognizing objects on a cooking counter.

Shape Reconstruction of Objects in Participating Media by Combining Photometric Stereo and Optical Thickness

Abstract: This paper proposes a method to reconstruct the 3D shape of objects in participating media. Shape reconstruction of objects in participating media, such as water, fog, and smoke, is difficult due to light scattering, which degrades image quality. While previous methods cope with this problem by removing the scattering components from images, the proposed method estimates optical thickness from images and uses it to recover the depth of the objects in participating media. With the proposed method, a detailed 3D shape is recovered using a photometric stereo technique that was designed to work in participating media. Three-dimensional global shapes that cannot be recovered by the photometric stereo technique, such as depth edges, are recovered from optical thickness. Experimental results with real images show that the proposed method correctly reconstructs the 3D shape of objects in participating media.