halcon-6.1-shape-matching.pdf

HALCON Application Note

How to Use Shape-Based Matching to

Find and Localize Objects

Provided Functionality

⊲ Finding objects starting based on a single model image

⊲ Localizing objects with subpixel accuracy

Typical Applications

⊲ Object recognition and localization

⊲ Intermediate machine vision steps, e.g., alignment of ROIs

find shape model, find shape models

find scaled shape model, find scaled shape models

write shape model, read shape model

clear shape model, clear all shape models

Copyright

c

 2002-2005 by MVTec Software GmbH, M¨unchen, Germany

MVTec Software GmbH

Overview

HALCON’s operators for shape-based matching enable you to find and localize objects based

on a single model image, i.e., from a model. This method is robust to noise, clutter, occlusion,

describes how to optimize the outcome of this phase by restricting the search space.

Shape-based matching is a powerful tool for various machine vision tasks, ranging from inter-

mediate image processing, e.g., to place ROIs automatically or to align them to a moving part,

to complex tasks, e.g., recognize and localize a part in a robot vision application. Examples can

be found in section 4.

Unless specified otherwise, the example programs can be found in the subdirectory

shape matching of the directory %HALCONROOT%\examples\application guide.

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or

transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise,

without prior written permission of the publisher.

Contents

1 A First Example

In this section we give a quick overview of the matching process. To follow the example ac-

tively, start the HDevelop program hdevelop\first example shape matching.dev, which

locates the print on an IC; the steps described below start after the initialization of the applica-

tion (press F5 once to reach this point).

reduce_domain (ModelImage, ROI, ImageROI)

After grabbing the so-called model image, i.e., a representative image of the object to find,

the first task is to create a region containing the object. In the example program, a rectangu-

lar region is created using the operator gen rectangle1; alternatively, you can draw the region

interactively using, e.g., draw rectangle1 or use a region that results from a previous segmen-

tation process. Then, an image containing just the selected region is created using the operator

reduce domain. The result is shown in figure 1.

inspect_shape_model (ImageROI, ShapeModelImages, ShapeModelRegions, 8, 30)

create_shape_model (ImageROI, NumLevels, 0, rad(360), 0, ’none’,

With the operator create shape model, the so-called model is created, i.e., the internal data

on the selected pyramid levels as shown in figure 1; thus, you can check whether the model

contains the relevant information to describe the object of interest.

When actually creating the model with the operator create shape model, you can specify

additional parameters besides NumLevels and Contrast: First of all, you can restrict the range

of angles the object can assume (parameters AngleStart and AngleExtent) and the angle

steps at which the model is created (AngleStep). With the help of the parameter Optimization

you can reduce the number of model points; this is useful in the case of very large models. The

find shape model. Note that if you use HALCON’s COM or C++ interface and call the oper-

ator via the classes HShapeModelX or HShapeModel, no handle is returned because the instance

of the class itself acts as your handle.

If not only the orientation but also the scale of the searched object is allowed to vary, you must

use the operator create scaled shape model to create the model; then, you can describe the

allowed range of scaling with three parameters similar to the range of angles.

for i := 1 to 20 by 1

grab_image (SearchImage, FGHandle)

find_shape_model (SearchImage, ModelID, 0, rad(360), 0.8, 1, 0.5,