Edge computing video data structuring method and system

Abstract

The invention discloses an edge computing video data structuring method and system and relates to the technical field of image processing. The method comprises steps of using image recognition techniques, obtaining a target image in the edge calculation video data, and reducing the target image according to a set reduction ratio P, generating S images and storing the S images, amplifying the target image according to a set amplification ratio Q; generating L images and storing the L images, calculating the target image, the S images and the L images by using a convolutional neural network; respectively obtaining categories, coordinate positions and category probabilities of the target image, each target image in the S images and each target image in the L images; depending on coordinate positions, respectively calculating the overlapping areas of the boundary frames where the target images of the same category are located in the target image, the S images and the L images, and when theoverlapping areas are greater than the set threshold, reserving the target images of which the category probability is greater than the set threshold, so that the efficiency and accuracy of detectingthe target images in the edge calculation video data are improved.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method and system for structuring edge computing video data. Background technique [0002] Video data structuring is to extract key information from the original video data by intelligently analyzing the original video data, and to describe the key information in text. At present, it mainly focuses on the structuring of people and vehicles. For example, the structuring of people includes extracting information such as gender, age, hairstyle, whether they wear glasses, whether they wear hats, clothing color, clothing type, etc. The structuring of cars includes extracting the license plate number, brand, model, and body color of the car and other information. [0003] In terms of realization of video structure, it is generally divided into two stages. The first stage first finds the target image from the video data, that is, finds a person or finds a car. The second ...

AI Technical Summary

Problems solved by technology

However, this processing method will cause the problem of low detection efficiency for small targets in high-definition large-size images

For example, in an image that is 1920 pixels wide and 1080 pixels high, if there is an object (such as a laptop) that is 60 pixels wide and 40 pixels high, after the image is scaled to 500 pixels wide and 500 pixels high, or even 300 pixels wide and 300 pixels When the pixel is high, the target image is difficult to detect when the size of the target image is only 10 pixels wide and 10 pixels high

[0005] Similarly, this processing method of setting a fixed input size also has the problem of low detection efficiency after the target image in the small-size image is enlarged.

For example, in an image that is 100 pixels wide and 100 pixels high, if there is a target image (such as a laptop) that is 60 pixels wide and 40 pixels high, when the image is scaled to 500 pixels wide and 500 pixels high, the target image is enlarged to 300 pixels wide and 200 pixels high, at this time the target image is also difficult to detect

Images