抄録

K-007
Efficient Scale Database Construction for Scale Filtering in SURF-based AR Application

In this research, multiple scalespaces of different sample resolutions are constructed to improve matching performance for SURF-based AR application. Using a simple ultrasonic range-finder, only appropriate scalespace is used for recognition process. Our prototype system is able to save at least 30 percent of computational resources and greatly increased the accuracy. To maintain the high performance, the database needs to be carefully constructed taking into account input quality and number of scalespace. Advantages, disadvantages of our proposal and possible improvements are also described.