Photogrammetry

28 May 2007

Photo-model-object

I have been thinking of different ways of explaining the different between a coplanarity and a collinearity approach to photogrammetry, or more accurately the difference between a RO/AO and a BA approach. Now, this is not something I have put a lot of thought into before and I must admit it's quite shameful, isn't it? Something so fundamental and I've neglected it completely. Until two days ago when I explained it like this to an LSGI3332 student:

1. Using Relative Orientation the focus is to create a geometrically stable and correct model by fulfilling (enforcing?) the coplanarity condition between two images. The focus is the model. Once the model is created (either by a coplanarity or a collinearity RO), if it's required to be put into object space a 3D conformal transformation is needed for Absolute Orientation. It's the model that is put into object space and the relationship between the cameras and the object are not specifically determined. This can me do by applying the AO to the RO result.

2. Using the Bundle Adjustment (simultaneous) approach, the focus is the relationship between the cameras and the object. Once this is achieved (or more correctly simultaneous with this operation) the model is defined with respect to the object space. Thus, once digitising begins, the model produced is in object coordinates.

Is that right?

28 November 2006

Food for thought

Conceptual problem?

I need some advice. How would you deal with this?:

"StereoMapper actually is the digital version of stereoplotter, taking topographic features from air photos. What it differs is its medium. On the hand, Aerial Photogrammetry extracts such kind of features directly from aircraft.

In general, StereoMapper is more convenient because we cannot always employ aircraft to film a region. Instead, we only need sit in front of computer and digitize the features we are longing for. It implies we can produce a topographic map whether it is stormy and sunny.

Also, the StereoMapper allows us to examine features very closely (to zoom in and out) so that we can collect and process image quite accurately. In contrast, aerial photos give us only an overview of the region although its quality is not bad. Its focus is fixed so that we cannot see its details accurately.

The most important thing is once we get aerial photos we have to do plainmetric mapping ourselves including radial triangulation and resection. This involves accurate hand and patient in order to map some features for only a region for a considerable time. Even mosaics and orthoimages, which are also photogrammetric products, need complex mathematical model to rectify photos and maintain their geometric relationship. But those vigorous calculation and photo rectification are absent in StereoMapper with just digitizing features that make up a topographic map.

Sometimes, it is not so advantageous to use stereo photogrammetry as features and sites change frequently because of new construction site and building projects. These change the landscape of a region. So we need to have aerial photos to tell us how change has been undertaken e.g. the effects of new buildings/roads upon the natural slope and vegetation.

Another method for generating a topographic map is that which is same as what we have done in practical 1. In practical 1, we need to use stereoscope to calculate some date which is the x, y, z value of the photo. As topographic map means that we use these values to produce.

The disadvantage of this method is that it needs to input and calculate all the data manually. However, StereoMapper is a computer program and we do all the process on the computer, we can output the data file to be ascii or text file and use excel to calculate the result. Also, it may still not a very accurate method when you compare it to analogue instruments which can magnify images and increase stability."

StereoMapper (SMP) is a product from 3DMapper and provides a reasonable environment for stereoscopic digitising. A companion module, StereoMaker (SMK), is used for the IO, EO and epipolar resampling processes required for SMP.

Choosing IO transformations.

I don't know how many times I talk about this with students and still read it in their reports:

"The RMS residuals tell us how accurate the inteior orientation is. The smaller the residuals, the more accurate is the result. In our practical the RMS residuals for the affine transformation is smaller than for the conformal transformation. That means the affine transformation more accurately describes the distortion of the photo. The residuals for the projective transformation are zero because there is no redundancy when there's only four fiducial marks. We cannot use the projective transformation because we cannot find any errors."

What's the problem with this type of argument?

15 November 2006

Suggestions

Ideas for information content, research projects, resorces, or anything else relating to photogrammetry.

Resources

Here I intend to collect URL's and other accessible material for your easy reference.

Cameras

All About Digital Cameras

World's Smallest Digital Cameras

Journals

Blackwell Synergy - PR

Organisations

Photogrammetry

3D Anaglyphs - Stereoscopic Pictures

Aerial Mapping Solutions

AeroMap US

AeroSys Consulting

Agfa Aerial Photography

AICON 3D Systems GmbH

Basic Course on producing maps from Aerial Photography

Camera Calibration

Camera Calibration Toolbox for Matlab

DIMAC - Digital Modular Aerial Camera

GeoSpace International

Geosystem

GPE

GSI

High Resolution Stereo Camera

ISTAR Mapping out in 3D

Optical Metrology Centre

PASCO

Paul Debevec Home Page

Stereoscopy.com - The World of 3D-Imaging!

Supresoft

SURE 340 Photogrammetry links

Triathlon

University of Melbourne

14 November 2006

Concepts and theories

Introduction

Photogrammetry is quite a simple technology. Just as our two eyes allow us to see in three dimensions, taking two photographs from slightly different positions will allow us to achieve the same result - stereoscopic vision. The real trick is how to take this 3D representation of the world and measure it. Photogrammetry then is the science of optimising photography for and the extraction of measurements, in 2 or 3 dimensions, from photographs.

In general, photogrammetry can be divided into two broad classes: those that mimic the human vision system and those that don't. The former goes by the name "stereo photogrammetry" and the latter "convergent photogrammetry". I'm sure you'll find many other classes too, but these are the two I will use and discuss here.

What is interesting is that the same basic principles can be applied to both types, and if I was pedantic enough I would say that stereo photogrammetry is just a special case of convergent photogrammetry. But whole textbooks have been written about each type (more on stereo than convergent) so I'll accept they can, and should, be treated differently. Once we look in detail at both types, it will become obvious why this evolution has happened.

Photogrammetry is a part of a broad science called remote sensing. Remote sensing encompasses all those techniques that can be used to obtain information about an object without touching it. There are many definitions to this field of study. In the world of Geomatics, some other examples are hydrographic surveying, satellite imaging and intersection survey.

Photogrammetry uses photographs either in hard copy or digital form to obtain the desired information. Some people think of a photograph only as a hard copy image. To save confusion, I will use the term image to represent any the item upon which we measure when doing photogrammetry.