Multi-resolution collaborative pedestrian identification method and related equipment

Abstract

The invention provides a multi-resolution collaborative pedestrian identification method and related equipment, and aims to solve the problem of pedestrian image resolution mismatch caused by distance change between a pedestrian and a camera in an actual monitoring scene. The multi-resolution collaborative pedestrian identification method specifically comprises the following steps: aiming at the problem of different scale space image identification values; proposing a multi-resolution image collaborative identification model, and jointly generating discrimination features by using the images before and after reconstruction; and a multi-feature weighted fusion mode is adopted, an optimal weight combination is obtained through calculation, and accurate expression of image features of any resolution is realized. Through the feature fusion calculation of the multi-resolution images, the identification values of the images with different resolutions are fully exerted, and the pedestrian identification capability in the resolution mismatch environment is improved.

Description

technical field [0001] The invention relates to the technical field of video analysis, in particular to a multi-resolution collaborative pedestrian identification method and related equipment. Background technique [0002] In a real application scenario, due to the randomness of the pedestrian's motion, the resolution of the pedestrian image captured by the camera is different due to the change of the distance between the pedestrian and the camera. Given a pedestrian image of arbitrary resolution, the task of finding pedestrian images with the same identity as the retrieved image from a retrieval library containing images of arbitrary resolution is defined as cross-resolution pedestrian retrieval. Since the distance between the object of interest and the camera is affected by the position of the monitoring point, the resolution of the images acquired by each monitoring point is also different. Direct comparison will seriously affect the recognition effect, making cross-resol...

AI Technical Summary

Problems solved by technology

However, there are two problems with this type of method: First, the image super-resolution is to improve the visual effect of the image rather than to improve the performance of pedestrian re-identification. The direct cascade of the super-resolution module and the re-identification module has the problem of poor compatibility. There are errors in the process of super-resolution, the greater the degree of super-resolution, the greater the error; second, these methods only use high-resolution images synthesized by low-resolution, and unify all images into a high-scale space, but the unified In different scale spaces, the noise of the image recovered after image super-resolution is also different, which will lead to different retrieval results

In other words, the reconstructed images in different scale spaces contain different image discriminative features. Traditional methods only focus on the features of high-scale space images, while ignoring the recognition value of images in other scale spaces.

Images