Digest Generation of Kindergarten Surveillance Video with Location Information and Visual Features

In this paper, we address the task of automatic digest generating of video data taken from kindergarten surveillance cameras. Our objective is extracting and merging video segments to recode children's daily life. In order to deal with mass video data efficiently, we jointly utilize location information and visual features to segment raw material videos. Our proposed method involves two steps. The first is to narrow down the searching space by analyzing the noisy RFID tag log which records kids' temporal location, while the second is to evaluate the quality of each unit video segment by combine multiple visual features. By merging high score unit segments, our method promises to efficiently provide high quality digest that could reflect kids' daily activities.