Book Title Recognition on 360VR Images for VR Tour of Library Building

This paper treats the optical character recognition (OCR) on 360VR images. The authors have already proposed a development framework for web-based VR tours, i.e., navigation VR tours of 360VR videos and walkthrough VR tours of 360VR images. They also proposed the extended development framework for web-based VR tours of 360VR videos based on OpenVSLAM(Open Visual SLAM: Simultaneous Localization and Mapping). Using OpenVSLAM, it is possible to generate a map consisting of several time-sequential locations and to extract their corresponding 360VR frame images from a navigation 360VR video. Furthermore, the authors introduced a keyword search function and a subtitle function into the proposed OpenVSLAM-Based development framework. The keyword search function is realized by OCR technology, i.e., Google Cloud Vision AI service, applied to extracted 360VR frame images and the subtitle function is realized by speech-to-text technology, i.e., Assembly AI service, applied to a narration sound file extracted from a navigation 360VR video. The authors have already developed web-based VR tours of the library building of the authors' university using the proposed framework. If the locations of books can be obtained by entering their book titles, the availability of the library building VR tour will become higher than ever. So, this paper treats the book title recognition on 360VR images. Although OCR results by Google Cloud Vision AI service are very good, the problem is that the service is not free. On the other hand, there are free OCR software, i.e., EasyOCR, TesseractOCR, and PaddleOCR. The authors performed OCR experiments for recognizing vertical book titles in Japanese kanji characters on 360VR images by the three OCR software and found that TesseractOCR has the best experimental results.