Distributed image analysis method and system and storage medium

Abstract

The invention discloses a distributed image analysis method, which comprises the following steps that a first server acquires a result set through image collision analysis, wherein the result set comprises an index image recording a result object, the index image corresponds to an object frequency, and the object frequency is used for indicating the shooting times of the result object; a pluralityof first servers respectively send result sets to a second server; the second server performs feature extraction on the index image in each result set to obtain a feature value of each index image; the second server performs image collision analysis on the extracted feature values of all the index images to obtain the confidence coefficient of each index image; when the second server determines that the confidence among the index images in the result set sent by the plurality of first servers is greater than or equal to a preset value, the sum of object frequencies corresponding to the indeximages is obtained.The embodiment of the invention further provides a distributed image analysis system and a storage medium. Calculation power of all levels of servers is fully utilized, and analysisefficiency is improved.

Description

technical field [0001] The invention relates to the technical field of image analysis, in particular to a distributed image analysis method, system and storage medium. Background technique [0002] With the advancement of big data and face technology, in video surveillance solutions, image collision analysis has penetrated into the security field, speeding up the detection efficiency of major and important cases, reducing the workload of case investigators in analyzing case information and improving case handling efficiency. In actual work, investigators can perform image collision analysis on the surveillance images captured by various network cameras; find people who frequently appear in a certain period of time and in a certain place, and the application scenario is such as the analysis of scalpers at a train station. [0003] In actual deployment and use scenarios, the video surveillance system is generally a multi-level domain composed of provinces (centers) and prefect...

AI Technical Summary

Problems solved by technology

[0004] In the prior art, since the artificial intelligence (AI) business algorithms of provincial centers and cities are provided by different manufacturers, the image collision analysis results of the city servers cannot be compatible with the provincial center servers. In order to solve this problem, the city The server can only upload local monitoring images to the provincial center server, and the provincial center server will perform unified reception, storage and image collision analysis on the monitoring images of various cities

As a result, the provincial center needs to temporarily save the business data uploaded by each city to perform an image collision analysis, and then recalculate and analyze it, which puts huge pressure on the computing and storage resources of the provincial center; and the system itself of each city It only has computing and analysis capabilities. Such a scheme cannot utilize the computing power resources of lower-level prefectures and cities, and only uses the prefecture-city system as a data storage system, resulting in a waste of computing power. A large number of picture streams and video streams need to be transmitted between them, which requires high network bandwidth and delay, and high construction costs. At the same time, due to the large capacity of monitoring images, it takes a long time for the provincial center to analyze image collisions

Images