Video retrieval method and system based on multiple image fusion

Abstract

The invention discloses a video retrieval method and system based on fusion of multiple images, belonging to the technical field of video retrieval. The method decodes the database video and divides the video shots in the offline indexing part to obtain multiple video shots, and single video shots The local feature sets of all key frames are pooled, and the reverse file index is established in units of video shots. In the online retrieval part, all query images are used, and the local features of all query images are pooled after quantization, and the quantized local feature pooling sets of all query images are obtained. According to the reverse file index, query images and videos The quantized local feature pooling set of shots is compared for similarity and video retrieval. And a video retrieval system based on fusion of multiple images is disclosed. The invention improves the recall rate of the video retrieval under the premise of ensuring the search efficiency.

Description

technical field [0001] The invention relates to the technical field of video retrieval, in particular to a video retrieval method and system based on fusion of multiple images. Background technique [0002] Automatic video data retrieval belongs to the problem of content-based video retrieval. Its purpose is to establish structure and index through computer processing, analysis and understanding of image / video content, so as to achieve convenient and effective image / video information acquisition. In recent years, researchers at home and abroad have done a lot of research on video retrieval systems. The intelligent video surveillance developed at present can detect, track and classify targets, and can monitor some unexpected abnormal events in real time. [0003] At present, the workflow of a general video retrieval system is as follows: figure 1 As shown, it mainly includes two parts: offline indexing and online retrieval. Offline indexing part: the first step is to extrac...

AI Technical Summary

Problems solved by technology

[0004] However, the drawbacks of the above-mentioned video retrieval technology are as follows: First, the recall rate is not high, because a query image is generally not enough to describe the target to be queried, especially for non-rigid or spatially topologically complex objects, and In practical applications, the omission of any traces of the monitoring target may lead to the loss of the monitoring target, so the recall rate is often more important than the precision rate in practical applications

The second is that the query efficiency is low. In the working process of the existing video retrieval technology, it is necessary to sort each key frame in the video, and then obtain the video sorting result by fusing the sorting results of the key frames. The number is far greater than the number of videos, so sorting key frames will result in slow target query speed and low resource utilization

Images