Method, device and system for automatically labeling target object in image

Abstract

The embodiment of the invention discloses a method, device and system for automatically labeling a target object in an image, and the method comprises the steps: obtaining an image training sample which comprises a plurality of images, each image is obtained through the shooting of the same target object, and the adjacent images have the same environmental feature points; Taking one image as a reference image, determining a reference coordinate system, and creating a three-dimensional space model based on the reference three-dimensional coordinate system; When the three-dimensional space modelis moved to the position where a target object is located in the reference image, determining position information of the target object in the reference three-dimensional coordinate system; And respectively mapping the three-dimensional space model to the image plane of each image according to the respective corresponding camera attitude information determined by the environmental feature pointsin each image. According to the embodiment of the invention, automatic image labeling can be carried out more accurately and effectively, and the universality of the method is improved.

Description

technical field [0001] The present application relates to the technical field of image processing, in particular to a method, device and system for automatically marking target objects in an image. Background technique [0002] In AR / VR and other related businesses, the use of machine learning methods to identify scenes / objects in images is widely used. In the process of machine learning, a large number of image training samples are needed, and the image training samples need to Label the target object. The so-called labeling means that the position of the target object in the image needs to be marked, so that the machine learning can perform feature extraction from the image of the target object for learning. [0003] In the prior art, there are mainly two types of labeling for image training samples, one is labeling based on two-dimensional images, and the other is labeling based on three-dimensional images of object CAD models. The so-called two-dimensional image annota...

AI Technical Summary

Problems solved by technology

However, its shortcomings are also very obvious, that is, under normal circumstances, the CAD model of the target object is usually provided by the manufacturer or designer of the target object, but if the manufacturer or designer cannot provide the CAD model, it will not be possible to use the above method to achieve automatic Labeling, and in practical applications, this phenomenon is very common, that is, it is difficult to obtain the CAD model of the target object, thus affecting the generality of this method

Secondly, even if the CAD model of the target object can be found, since the tracking of the target object usually relies on enough feature points on the object, when the object itself is pure color, highly reflective or transparent, model-based tracking will It cannot guarantee its sufficient accuracy, which will affect the effect of automatic labeling

Images