Method and device for enhancing accuracy of image analysis algorithm

Abstract

The invention provides a method and a device for enhancing the accuracy of an image analysis algorithm. The method comprises the steps of performing video decoding treatment on an acquired video stream to acquire a plurality of video frames; performing denoising treatment on each video frame to acquire a plurality of denoised video frame; performing image analysis on each denoised video frame to acquire a plurality of analysis result; successively storing the plurality of analysis results into an array, and when the quantity of the analysis results reaches the length of the array, averaging the analysis results to acquire a plurality of average value analysis results; successively sending the plurality of average value analysis results into a sliding window, and when the quantity of average value analysis results reaches the length of the sliding window, correcting the average value analysis result of the middle position of the sliding window until all the average value analysis results are completely processed. According to the scheme, the video image is firstly denoised to optimize image quality, so that the definition is enhanced, then a sliding window algorithm is combined to further optimize the analysis result, so that the accuracy is improved.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method and device for improving the accuracy of an image analysis algorithm. Background technique [0002] The trial review system is a set of systems that conduct normative inspections on the behavior of judges, clerks and parties involved in the trial process. This system combines a number of artificial intelligence technologies, and image analysis technology can be used to To analyze and discover the irregular behavior of judges, clerks and parties in the trial process, and remind them in time to achieve the purpose of standardizing the order of the trial. [0003] However, due to some reasons, the actual application effect is not very ideal. mainly includes: [0004] 1. Since the court trial involves multiple parties, most of the court trial videos use screen synthesis technology to display multiple parties in different split screens in the same video, which lead...

AI Technical Summary

Problems solved by technology

[0004] 1. Since the court trial involves multiple parties, most of the court trial videos use screen synthesis technology to display multiple parties in different split screens in the same video, which leads to a very small proportion of the screen occupied by the judge and the parties, and the screen is clear not enough;

[0005] 2. Limited by the network conditions of the court and the actual lighting effect of the court, there are many noises in the picture, and the quality and clarity are not very good;

[0006] Due to the influence of some of the above objective reasons, when the image analysis algorithm of the court review system checks these videos with low quality and clarity, it is easy to have wrong analysis results, false positives or false positives; resulting in court users The feeling and experience of the system is not good, which leads to the reduction of purchase intention

Images