Recognition of Browsing a Book Using Smartwatches Toward Inferring Interest

For physical stores, it would be useful if they can know to what degree customers are interested in each item placed in the store before purchasing. There have been many studies on relationship between a person's behavior and interest to targets mainly focusing on their gazes. However, considering cases in which persons move around a store, keeping track of gazes of them using multiple camera devices and so on will be costly. This paper focuses on cases of browsing several books in a bookstore and presents a novel scheme to recognize a motion of browsing a book using one of popular wearable devices, a smartwatch. Time series of gyro sensor and acceleration sensors are analyzed to recognize time length of browsing and the number of times a page turned during browsing. Results of experiments show it can achieve F value over 0.8 regarding recognition of browsing time sections, and also show that the number of a page turned is recognized with high accuracy for a motion of turning a single page at a time.