Quick filtration system facing large scale video analysis

Abstract

The invention discloses a quick filtration system facing large scale video analysis. The quick filtration system facing large scale video analysis includes a multistage filter, a full characteristic network model, and a global feedback queue module, wherein the multistage filter is used for filtering background frames, frames without target objects, and frames that the target object number is lessthan the value threshold Number of Object, in video streams, so as to obtain the detected frames in the video streams; the full characteristic network model is used for detecting and analyzing the detected frames; the global feedback queue module is used for coordinating the processing rate of each filter in the multistage filter; and a queue is added between the front and back two continuous filters of the multistage filter, so as to enable the multistage filter to perform asynchronous parallel execution in a mode of a streamline. Through cascade connection of the plurality of filters, the quick filtration system facing large scale video analysis can greatly reduce the number of the video frames requiring analysis of the full characteristic model, and through the asynchronous parallel cascade structure, can improve the number of the video streams which can be detected by the system in real time in the same hardware environment.

Description

technical field [0001] The invention belongs to the field of big data systems and analysis, and more specifically relates to a fast filtering system for large-scale video analysis. Background technique [0002] Surveillance cameras are everywhere around us. For these huge amounts of video data, the traditional approach is to collect them in an operation center for manual observation. This approach is not only error-prone, but also expensive. By using the method of machine learning, automatic video analysis based on object detection technology can reduce human intervention to a great extent, so it has received extensive attention. The analysis of surveillance video can be divided into two types: online real-time analysis and offline post-event analysis. Under the same hardware environment, online analysis requires expanding the number of video streams as much as possible, and offline analysis requires searching and querying video files as quickly as possible. [0003] How...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the above defects or improvement needs of the prior art, the present invention provides a fast filtering system oriented to large-scale video analysis, thereby solving the problem that the prior art has a slow detection and analysis speed and cannot perform fast filtering in a limited hardware environment. Technical issues of offline analysis and online analysis of large-scale video streams

Images