Image labeling method and device based on three-dimensional reconstruction

Abstract

The embodiment of the invention discloses an image annotation method and device based on three-dimensional reconstruction, a feature map can be constructed according to an initial image acquired by image acquisition equipment in an automatic driving vehicle by utilizing a three-dimensional reconstruction principle, and one-time operation, namely one-time annotation, on the feature map is manually performed, so that the image annotation efficiency is improved. And a plurality of two-dimensional images can be projected to obtain annotated images, so that the annotation efficiency can be greatly improved. Moreover, by means of high-precision three-dimensional reconstruction, one-time manual labeling can be accurately projected to a plurality of two-dimensional images, labeling consistency is achieved, wrong labeling is avoided, and therefore the accuracy of image labeling can be improved. Besides, when the trained target detection model is used for element perception, the perceived current target element can be projected on the two-dimensional image based on the three-dimensional reconstruction result, and the perception result of the target detection model is compared with the projection result to determine the perception result of the target detection model. Therefore, the accuracy of the model can be evaluated.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to an image labeling method and device based on three-dimensional reconstruction. Background technique [0002] Image recognition based on deep learning requires a large number of sample images to train the model, and before model training, it is necessary to first label each sample image to obtain the labeling results. [0003] The known sample labeling methods usually use a pre-trained model to assist, that is, use the pre-trained model to generate a label frame on the image to be labeled, and then manually adjust it. However, the stability of the annotation frame generated by the pre-training model is poor, there may be errors and omissions, and manual verification is required one by one, which leads to high labor costs and low annotation efficiency when the number of samples is large. Therefore, in order to improve the efficiency of image annotation, an image a...

AI Technical Summary

Problems solved by technology

However, the stability of the annotation frame generated by the pre-training model is poor, there will be errors and omissions, etc., and manual verification is required one by one, resulting in high labor costs and low annotation efficiency when the number of samples is large.

Images