Searching Top-K Similar Moving Videos

Abstract

The application of sensors enables mobile devices to generate amounts of content-aware data, such as trajectory, gyro, and video data. Moving video is an emerging new type of moving object that can provide a potential source for geo-referenced applications. Measuring the similarity of moving videos is widely used in traffic management, tourist recommendations, and location-based advertising. Our prior work proposed two similarity measures, the Largest Common View Subsequences can accurately calculate similar moving videos, and the View Vector Subsequences can fast calculate similar moving videos. In this paper, we proposed the searching top-k similar moving videos (K-SSMV). First, we give the problem definition of searching top-k similar moving videos. Then, we illustrated the strategy of searching for moving videos. Specifically, we sorted the video pairs by the most similar videos. Since the similar videos would be less than k, we first calculated the similarity of moving videos by the Largest Common View Subsequences algorithm, if the number of similar videos is less than k, we will adopt the View Vector Subsequences algorithm to compute the similarity of moving videos. Next, we mixed the candidate videos calculated by the Largest Common View Subsequences and the View Vector Subsequences algorithms, the searching top-k similar moving videos algorithm picked the top-k similar videos from candidate videos. Finally, we evaluated the performance of our proposed algorithms on the accuracy and computational cost. The experiments verified that our algorithms can efficiently search top-k similar moving videos.