Automated surveying of real world objects

Patent metadata

Field	Value
Publication number	US-2019178643-A1
Application number	US-201816200326-A
Country	US
Kind code	A1
Filing date	Nov 26, 2018
Priority date	Dec 11, 2017
Publication date	Jun 13, 2019
Grant date	—

Abstract

A method for automatic surveying of a real word object by a surveying or metrology instrument. It comprises an acquiring of a picture by the surveying or metrology instrument, which picture is comprising the real world object to be surveyed, and an automatic detecting and classifying of the detected real world object to be surveyed in the picture, and also an automatic surveying of the detected real world object by the surveying or metrology instrument for determining a location and/or a geometrical property of the real world object. The automatic detecting and classifying is done with a classifier based on data acquired by machine learning, which machine learning is based on machine learning training data derived from a virtual digital 3d-model which is representing the real world object and which comprises meta information of this 3d-model.

First claim

What is claimed is:

1 . A method for automatic surveying or measuring of a real world object by a surveying or metrology instrument, the method comprising:
acquiring a picture by the surveying or metrology instrument, which picture comprises the real world object which is to be surveyed or measured; automatic detecting and classifying of the real world object within the picture; and automated measuring of the detected and classified real world object by the surveying or metrology instrument for determining a location or a geometrical property of the real world object, wherein: the automatic detecting and classifying is done with a classifier based on a classification-model acquired by machine learning, the machine learning being based on machine learning training data comprising a virtual digital three-dimensional (3D) model, representing the real world object and comprising meta information of this 3D model.

2 . The method according to claim 1 , wherein the meta information is comprising at least a class-, type- and name-identifier for the real world object.

3 . The method according to claim 1 , wherein the machine learning training data is comprising a plurality of numerically rendered images derived from the 3D model.

4 . The method according to claim 3 , further comprising:
rendering the numerically rendered images from the 3D model with a virtually simulating environmental conditions of the 3D model.

5 . The method according to claim 4 , wherein the environmental conditions comprise at least one of:
a viewing angle, an illumination, a shading, a foreground, a background, a fog, or a scale.

6 . The method according to claim 1 , wherein the metadata comprises an orientation information of the object in the virtual training data.

7 . The method according to claim 1 , wherein the real world picture comprises 3D information and the numerically rendered images comprise 3D information.

8 . The method according to claim 1 , wherein the 3D model and the resulting learning data comprise the meta information with at least one surveying-feature, wherein the surveying or metrology instrument is automatically measuring at least one of the at least one surveying features.

9 . The method according to claim 1 , wherein the at least one surveying-feature comprises at least one of an axis, symmetry, dimension, distance, angle, diameter, depth, relative proportions, geometric dependency, scalability.

10 . The method according to claim 3 , wherein the rendered images depict a generic geometrical feature from the 3D model, representing a generic geometrical feature of a plurality of real world objects and comprises meta information on surveying this generic geometrical feature, in particular wherein an untrained real world object is automatically measurable based on the automatic detecting and classifying of one or more of the generic geometrical features.

11 . A method of machine learning a surveying or metrology instrument, for the surveying instrument to establish an automatic detecting of a real world object in camera pictures taken by the instrument, the method comprising:
deriving of training data from a 3D model representing the real world object and comprising meta information of the real world object it represents; and training an object detector/classifier by machine learning on said training data.

12 . A surveying instrument built for carrying out the method according to claim 1 , comprising a camera for taking real world pictures and comprising a classifier with a machine learned neural network being trained on numerically rendered images which are derived from a virtual 3D model.

13 . A metrology instrument built for carrying out the method according to claim 1 , comprising a camera for taking real world pictures and comprising a classifier with a machine learned neural network being trained on numerically rendered images which are derived from a virtual 3d-model.

14 . The instrument according to claim 12 , wherein providing an autonomous measurement mode, which automatically measures geometrical features of a real world object automatically identified by the classifier in the real world pictures, according to trained meta information from the virtual 3d-model.

15 . A computer program product comprising program code stored on a machine-readable medium, for executing a method according to claim 1 .

Assignees

• Hexagon Technology Ct Gmbh

Inventors

• Metzler Bernhard
• Pryczek Michal Pawel

Classifications

• G01C11/025 Primary

  by scanning the object · CPC title

• G01C11/04 Primary

  Interpretation of pictures · CPC title

• G06V10/774

  Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06N3/045

  Combinations of networks · CPC title