부산대학교 도서관

This paper describes a high speed image retrieval method executable on smartphones in real-time. This method will enable us to develop an application of smartphones which assists the visually impaired to shoot a video of an object and to distinguish that object without using the internet. Taking a video makes a visually impaired user to put the object in the camera frame correctly, and the smartphone without the internet protects the user's privacy. Our method is constructed by ORB local features and a data structure of binary decision tree. Under the experimental environment of 3.1 GHz Intel Core i7 platform with Windows 10, our system performs 54.5 ms processing time and 96.1% accuracy. Furthermore, our method achieves 30 fps processing time on a smartphone.