Developing a Real-World Task Support System Requiring Recognizing Many Symbols by Using Machine Learning and AR Technologies
鈴木涼介, 大囿忠親, 新谷虎松


AR 技術を利用した実空間における作業支援システムの構築には,機械学習の利用が欠かせない.機械学習の適用において,学習済モデルの構築が容易ではなく,支援が必要である.本研究では,多様な記号の認識が必要な実空間における作業支援に関する検討の一環として,麻雀支援のためのAR 技術に基づく支援システムを試作した.麻雀支援は,多様な図形の認識や,実環境からのコンテキストの認識が必要であり,挑戦的な課題である.本稿では,AR 技術に基づく麻雀支援における機械学習の適用事例を紹介する.
We use machine learning and AR(Augmented Reality) technologies to develop a real-world task support system requiring recognizing many symbols. The system supports to make a learning model for recognizing the symbols. We have developed a Mah-jong support system as a concrete application of the system. Mah-jong is a popular indoor game in Japan. We realized the mechanism for recognizing 136 pieces shaped like small tiles, each identified by a kanji character inscribed on the face by using the machine learning. The system can display the score on the winning combination of tiles by using the AR technology. We show how the system can be effectively used and illustrate its practically using case studies.