A Novel Video Retrieval Method Based on Object Detection Using Deep Learning

Abstract

The recent research in computer vision is focused on videos. Image and video data has been increased drastically in the last decade. This has motivated researchers to come up with different methods for image and video understanding and other applications like action recognition from videos, video retrieval, video understanding, and video summarization. This article proposes a novel and efficient method for video retrieval by using the deep learning approach. Instead of considering low-level features, videos could be best represented in terms of the high-level features for achieving efficient video retrieval. The idea of the proposed research work is novel, where objects present in the query video are used as features and are used to match against all other videos in the database. Here, object detection is based on YOLOv3, which is the current state-of-the-art method for object detection from videos. This method is tested against YouTube action dataset. It was found that the proposed method has obtained comparable results as the other state-of-the art video retrieval methods.KeywordsVideo retrievalDeep learningYolov3Object detection