Video Lifelog Retrieval System for Ambiguous Search Queries
Mayo Morimoto, Masashi Fujitsuka, Sawako Mikami, Yosuke Motohashi · 2020
The fifth-generation mobile network (5G) will soon become popular, and data utilization will increase. Although there is a large volume of video data, making it difficult to manage and use such data in our daily lives, there is huge potential with such data. We now often take photos using smartphones, but in future recording our lives by wearing video cameras would become popular. Since such data have a large amount of information, a video lifelog retrieval system is necessary. Although many studies on video retrieval have been conducted on some domains, e.g. TV programs, movies and so on, there have been few conducted on lifelogs. One of the main differences between them is regarding user queries and the fact that our memories are often ambiguous. We propose a video lifelog retrieval system that takes advantage of ambiguous search queries. We discuss the effectiveness of our system from our evaluation involving our own lifelog data we collected. We also discuss the characteristics of lifelog data and problems of our system.