For the purpose of establishing a digital twin society, this paper proposes an object recognition network architecture using high-definition (HD) camera images. HD camera is installed to the edge computing resource in Road-Side Unit (RSU). The edge computing resource detects vehicles by object detection and Re-ID function based on deep learning. Its key feature is to reduce the amount of network data traffic by extracting vehicle images and directly transferring them to neighboring RSUs, i.e. vehicle identification and tracking can be achieved in the localized backhaul network. This paper experimentally verifies its fundamental feasibility using outdoor testbed field. The proposed framework can significantly reduce the data traffic by more than 90% while maintaining vehicle Re-ID accuracy.

Consistency is one of the keys to maintainable source code and hence a successful software project. We propose a novel method of extracting the intent of programmers from source code of a large project (~ 300 kLOC) and checking the semantic consistency of its variable names. Our system learns a project-specific naming convention for variables based on its role solely from source code, and suggest alternatives when it violates its internal consistency. The system can also show the reasoning why a certain variable should be named in a specific way. The system does not rely on any external knowledge. We applied our method to 12 open-source projects and evaluated its results with human reviewers. Our system proposed alternative variable names for 416 out of 1080 (39%) instances that are considered better than ones originally used by the developers. Based on the results, we created patches to correct the inconsistent names and sent them to its developers. Three open-source projects adopted it.

We investigated how programmers express high-level concepts such as path names and coordinates using primitive data types. While relying too much on primitive data types is sometimes criticized as a bad smell, it is still a common practice among programmers. We propose a novel way to accurately identify expressions for certain predefined concepts by examining API calls. We defined twelve conceptual types used in the Java Standard API. We then obtained expressions for each conceptual type from 26 open source projects. Based on the expressions obtained, we trained a decision tree-based classifier. It achieved 83 % F -score for correctly predicting the conceptual type for a given expression. Our result indicates that it is possible to infer a conceptual type from a source code reasonably well once enough examples are given. The obtained classifier can be used for potential bug detection, test case generation and documentation.

Head-Mounted Displays, or HMDs, are rapidly spreading in recent years. However, existing text input methods for HMDs have several problems, hampering their further popularization. In this paper, we focus on the two problems with existing text input methods: (1)the difficulty of its setup and (2) the space needed for its operation. To address these problems, we propose a novel text input system for HMDs. Our system (1) builds on top of a commodity camera device, Leap Motion, and (2) enables effective input in a small physical/virtual space, with a virtual 12-key keyboard. Our experimental results show the advantage of our approach over existing hand-tracking text input systems; especially our method indicates the effectiveness for Japanese text input, with a small virtual keyboard. Based on the experimental results, we believe that our text-input system is effective for any language in HMD environments, if we have a 12-key keyboard tuned for each language (as well as Japanese).

We are trying to find source code comments that help programmers understand a nontrivial part of source code. One of such examples would be explaining to assign a zero as a way to "clear" a buffer. Such comments are invaluable to programmers and identifying them correctly would be of great help. Toward this goal, we developed a method to discover explanatory code comments in a source code. We first propose 12 distinct categories of code comments. We then developed a decision-tree based classifier that can identify explanatory comments with 60% precision and 80% recall. We analyzed 2,000 GitHub projects that are written in two languages: Java and Python. This task is novel in that it focuses on a microscopic comment ("local comment") within a method or function, in contrast to the prior efforts that focused on API- or method-level comments. We also investigated how different category of comments is used in different projects. Our key finding is that there are two dominant types of comments: preconditional and postconditional. Our findings also suggest that many English code comments have a certain grammatical structure that are consistent across different projects.