A surveillance video multi-target classification and retrieval method and system based on deep learning

Abstract

The invention discloses a monitoring video multi-target classification retrieval method and system based on deep learning; extracting the target to be retrieved: inputting the image to be retrieved, selecting the target to be retrieved through automatic detection or manual extraction, and specifying its category; extracting Depth features: use the deep learning feature extraction model to extract the depth features of the target to be retrieved; target retrieval: compare the depth features of the target to be retrieved with the depth features of similar targets in the retrieval database according to the category of the target to be retrieved, and obtain the The most similar target; search results display: search results are displayed in order of similarity from high to low. The depth features acquired by the deep learning feature extraction module are used as the benchmark for matching the target to be retrieved with similar targets in the retrieval database. Because the depth feature has the dual advantages of strong expressiveness and low dimensionality, the accuracy of the retrieval results is greatly improved.

Description

technical field [0001] The invention relates to a video target retrieval technology, in particular to a deep learning-based monitoring video multi-target classification retrieval method and system. Background technique [0002] In order to create a safer living environment and meet the increasing security needs of the people, the state vigorously promotes the construction of safe cities, the core basis of which is the huge number of video surveillance equipment deployed in every corner of the city. In addition to the important application scenario of urban security monitoring, as the most effective way in the field of security, video surveillance equipment has been widely used in areas closely related to daily life such as airports, stations, banks, shops and communities. Therefore, in recent years, the number of video surveillance devices has increased dramatically, and a huge amount of video surveillance devices are generating massive amounts of surveillance video data eve...

AI Technical Summary

Problems solved by technology

[0003] Security monitoring video has problems such as complex scene environment, dense targets, and various resolutions. Especially when multi-target classification retrieval is performed, the classification and extraction of targets in surveillance video is more susceptible to factors such as background noise and target occlusion than general target extraction. Shallow features are also difficult to accurately express multiple types of targets with large differences in appearance using unified features. In recent years, the development of deep learning can just solve the above problems

Images