Target labeling method and device based on 3D virtual driving scene

Abstract

The invention provides a target labeling method and device based on a 3D virtual driving scene, and the method comprises the steps: establishing the 3D virtual driving scene, and setting a to-be-labeled virtual scene element at a preset position of the virtual driving scene; obtaining spatial three-dimensional coordinates of four vertexes of a minimum bounding rectangle tangent to the front surfaces of the virtual scene elements, performing virtual driving shooting through a virtual vehicle carrying a virtual camera according to a virtual driving route, and defining a picture containing the to-be-labeled virtual scene elements in a picture as a key background picture, and labeling the virtual scene elements in the minimum bounding rectangle in the key background picture in batches. According to the method, batch labeling and batch exporting of the targets are realized by means of the 3D virtual scene, and the labeling efficiency and precision are improved.

Description

technical field [0001] The invention relates to the field of automobile intelligent driving, in particular to a method and device for marking objects based on a 3D virtual driving scene. Background technique [0002] Currently, most data annotations are for video images. The traditional annotation method generally needs to extract each frame of image first, and give the coordinate position marking information of the key attention factors in the image, which is inefficient and consumes high system CPU resources. [0003] In addition, the patent CN106791937A proposes a method and system for labeling video images. A canvas is created, and the pixels of the canvas correspond to the pixels of the video image to be labeled. According to the received labeling command and the corresponding relationship between the pixels, the label is drawn at the specified position on the canvas. content, convert the format of the canvas to obtain the first labeled image, and assign the pixel valu...

AI Technical Summary

Problems solved by technology

At present, the data annotation of driving images is to manually annotate a large number of captured road conditions images one by one, and the annotated data is used for the learning of the vehicle camera. This method consumes a lot of manpower and material resources, and the efficiency is low.

[0005] The current data annotation is to process two-dimensional images, and there are few achievements in the specific area annotation in the 3D virtual scene that reconstructs the objective world, and the existing annotation methods are extremely quantitative in terms of annotation accuracy and annotation efficiency

Images