Method and device for recognizing a known object in a field of view of a three-dimensional machine vision system

Patent metadata

Field	Value
Publication number	US-9483707-B2
Application number	US-201514614014-A
Country	US
Kind code	B2
Filing date	Feb 4, 2015
Priority date	Feb 4, 2015
Publication date	Nov 1, 2016
Grant date	Nov 1, 2016

Abstract

A method for recognizing a known object in a field of view of a three-dimensional (3D) vision system includes capturing a bitmap image file of the field of view and executing a first feature extraction to generate appearance features including a color histogram and an edge direction histogram. A reduced search space is identified based upon the appearance features including the color histogram and the edge direction histogram. A second feature extraction of the bitmap image file is executed to identify geometric features including a viewpoint feature histogram and an ensemble of shape functions. The appearance features and the geometric features are fused with contents of the reduced search space. The fused appearance features, geometric features and contents of the reduced search space are compared with a plurality of predetermined partial views of a plurality of known objects to recognize one of the known objects.

First claim

The invention claimed is:

1. A method for recognizing a known object in a field of view of a machine vision system, comprising:
capturing a bitmap image file of the field of view employing the machine vision system;
executing a first feature extraction of the bitmap image file to generate appearance features including a color histogram and an edge direction histogram, wherein the color histogram includes the bitmap image file quantized for hue and color saturation, and wherein the edge direction histogram includes the bitmap image file quantized for frequency and directionality of brightness changes of an object captured in the bitmap image file;
identifying a reduced search space based upon the appearance features, including comparing the color histogram and the edge direction histogram with features of partial views of the known objects, wherein the reduced search space includes a portion of data contained in the bitmap image file;
executing a second feature extraction of the bitmap image file to identify geometric features including a viewpoint feature histogram and an ensemble of shape functions associated with the bitmap image file, wherein the viewpoint feature histogram is determined by encoding statistics of a shape of an input point cloud associated with the bitmap image file by accumulating geometric relations between all point pairs;
fusing the appearance features and the geometric features with the reduced search space;
wherein said fusing includes comparing the fused appearance features, geometric features and the reduced search space with a plurality of predetermined partial views of a plurality of known objects; and
recognizing, by the controller, one of the known objects based upon the comparing.

2. The method of claim 1 , wherein accumulating geometric relations between all point pairs of the input point cloud comprises accumulating a four dimensional histogram between each of the point pairs.

3. A machine-executed vision-based object recognition process, comprising:
capturing a plurality of predetermined partial views for each of a plurality of known objects;
capturing, via a machine vision system, a bitmap image file including a 3D image of the field of view including at least one of the known objects employing the machine vision system;
executing, by a controller, a two-dimensional (2D) feature extraction of the bitmap image file to generate 2D appearance features including a color histogram and an edge direction histogram, wherein the color histogram includes the bitmap image file quantized for hue and color saturation, and wherein the edge direction histogram includes the bitmap image file quantized for frequency and directionality of brightness changes of an object captured in the bitmap image file;
identifying a reduced search space based upon the appearance features, including comparing the color histogram and the edge direction histogram with features of partial views of the known objects, wherein the reduced search space includes a portion of data contained in the bitmap image file;
executing a 3D feature extraction of the bitmap image file to identify geometric features including a viewpoint feature histogram and an ensemble of shape functions associated with the bitmap image file, wherein the viewpoint feature histogram is determined by encoding statistics of a shape of an input point cloud associated with the bitmap image file by accumulating geometric relations between all point pairs;
fusing the appearance features and the geometric features with the reduced search space;
wherein said fusing includes comparing the fused appearance features, geometric features and contents of the reduced search space with the plurality of predetermined partial views of the plurality of known objects; and
recognizing, by the controller, one of the known objects based upon the comparing.

4. A machine vision device, comprising:
an image detector signally connected to an encoder signally connected to a controller; and
the controller executing a vision-based object recognition process, including the following steps:
capturing, via the image detector and encoder, a bitmap image file including a 3D image of the field of view including at least one of the known objects employing the machine vision system;
executing a two-dimensional (2D) feature extraction of the bitmap image file to generate 2D appearance features including a color histogram and an edge direction histogram, wherein the color histogram includes the bitmap image file quantized for hue and color saturation, and wherein the edge direction histogram includes the bitmap image file quantized for frequency and directionality of brightness changes of an object captured in the bitmap image file;
identifying a reduced search space based upon the 2D appearance features, including comparing the color histogram and the edge direction histogram with features of partial views of the known objects, wherein the reduced search space includes a portion of data contained in the bitmap image file;
executing a 3D feature extraction of the bitmap image file to identify geometric features including a viewpoint feature histogram and an ensemble of shape functions associated with the bitmap image file, wherein the viewpoint feature histogram is determined by encoding statistics of a shape of an input point cloud associated with the bitmap image file by accumulating geometric relations between all point pairs;
fusing the appearance features and the geometric features with the reduced search space;
wherein the comparing includes comparing the fused appearance features, geometric features and the reduced search space with a plurality of predetermined partial views of the plurality of known objects; and

recognizing one of the known objects based upon the comparing.

5. The method of claim 2 , wherein the four dimensional histogram is related to a first angle, a second angle, a third angle and a distance between the points of each of the point pairs.

6. The method of claim 5 , wherein the four dimensional histogram related to the first angle, the second angle, the third angle and the distance between the points of the point pair include a first angle θ, a cosine of the second angle α, a cosine of the third angle φ, and a distance between the points of the point pair.

7. The method of claim 6 , wherein the first angle θ, second angle α and third angle φ represent corresponding pairwise pan, tilt and yaw angles that are determined for every pair of points of the point pair.

8. The method of claim 6 , wherein the distance between the points of the point pair is normalized.

Assignees

• Gm Global Tech Operations Llc

Inventors

• Chen Zhichao

Classifications

• G06V10/806

  of extracted features · CPC title

• H04N13/271 Primary

  wherein the generated image signals comprise depth maps or disparity maps · CPC title

• G06F18/253

  of extracted features · CPC title

• G06F18/2148

  characterised by the process organisation or structure, e.g. boosting cascade · CPC title

• G06F18/2413

  based on distances to training or reference patterns · CPC title