A Parallel Top-N Video Big Data Retrieval Method Based on Multi-features

Abstract

At present, a large amount of video data has been generated at an amazing speed through social media, industrial Internet and high-tech visual devices. When videos are retrieved, since the traditional video retrieval method often requires a lot of execution time, and takes a little account of the multi-source heterogeneous characteristics of videos, it is difficult to fully meet the needs of people’s daily life. Therefore, it is necessary to study a retrieval method for video big data to address the above issues. At present, there are few studies on video big data retrieval methods. The existing studies have an issue that the retrieval efficiency of video big data is low due to the sorting of all data nodes in the shuffle stage. To address this issue, this paper proposes a parallel top-N video big data retrieval method based on multi-features. First, the low-level and high-level semantic features of each video are extracted distributed. Second, a local sensitive hashing algorithm is introduced to partition and compress the video features to improve the efficiency of similarity matching for video data. Finally, a parallel top-N algorithm is constructed to quickly sort and summarize the similarity matching results of videos, so as to detect the approximate video data in the video big data. In this paper, experiments are carried out on Stanford I2V dataset. The results show that compared with the existing methods, the proposed method can effectively improve the efficiency of video big data retrieval while ensuring the accuracy of video big data retrieval.