PE&RS May 2001

Letters to the Editor

Dear Editor,

The article titled, “An Evaluation of an Off-the-Shelf Digital Close-Range Photogrammetric Software Package,” (PE&RS, February 2001) describes the product PhotoModeler which is developed by my company, Eos Systems Inc. There are a number of statements in this article that may lead to misinterpretation of PhotoModeler’s capabilities.

First, I want to state that Eos Systems did not see nor review this article before its publication. The ack-nowledgement given to Eos Systems at the end of the article does not represent acceptance or any pre-knowledge of the article’s contents on our part.

The article’s main focus was the evaluation of PhotoModeler’s accuracy with non-metric cameras. The definition of a non-metric camera is hence important. A metric camera is a film or digital camera designed specifically for metrology and a non-metric camera has other design goals. The strong interest in non-metric cameras, expressed in the article and elsewhere, is valid because these cameras are accessible and affordable. PhotoModeler handles images from metric cameras as well, but the focus of the article was non-metric cameras so we will limit our discussion to them.

The article did not clearly state that it focused on only one possible method of using PhotoModeler. The article’s study used images from non-metric cameras of unknown parameterization. This specific use (where the internal camera parameters are not known) is just a subset of non-metric camera use – we call this Inverse Camera processing in PhotoModeler.

Inverse Camera processing requires the program to reconstruct both the camera’s parameters as well as the scene’s 3D geometry. The use of Inverse Camera is quite common in some fields, such as forensics and accident reconstruction, but it is not at all typical in other fields, such as architectural or industrial measurement. In these other fields one would use a high quality pre-calibrated digital camera instead. In this latter case Photo-Modeler has shown accuracy of up to 1 part in 20,000 (see the Accuracy Studies page at www.photomodeler.com). An even higher accuracy is possible with metric cameras.

The article’s achievement of 1 part in 1700 accuracy for a non-metric Inverse Camera project is quite reasonable, but it is not indicative of PhotoModeler’s best, or even typical, performance.

The article’s statement that PhotoModeler is “relatively low accuracy”, in reference to the Fraser article, is taken out of context. The Fraser article is written from the standpoint of industrial metrology and makes a separation of close-range systems into low accuracy (less than 1 part in 5000), high accuracy (greater than 1 part in 50,000), and medium accuracy (between these). While most Inverse Camera PhotoModeler projects would fall into this “low accuracy” category, it is quite easy to obtain “medium accuracy” with calibrated non-metric cameras and PhotoModeler.

The article states that the Photo-Modeler’s Camera Calibrator is “of trivial value”. This may be true for unstable non-metric film cameras, but it is not true for the majority of Photo-Modeler customers that use modern non-metric digital cameras or metric film cameras.

The article also describes a “Pseudo Camera” method to overcome a perceived deficiency in PhotoModeler by the author. We feel that the Pseudo Camera method is not appropriate for the majority of PhotoModeler users and indeed PhotoModeler’s implementation is the correct one. It is possible the author did not have his camera data initialized correctly, which led him to see this non-existent deficiency.

PhotoModeler is a widely used, award-winning and accurate close-range photogrammetric package.

Alan Walford
President
Eos Systems Inc.

Response
The statements in Mr. Walford’s letter, which are found generally objective and understandable from the angle of the company concerned, reflect the understanding of the paper and the relevant photogrammetric issues, likewise, our paper reflects our understanding of our test results and the relevant photogrammetric issues in interpreting, analyzing and summarizing the results. We believe that both the paper and the letter shall be helpful to the readers, while the readers would find that the combination could be more helpful.

We are in fact glad to see that PhotoModeler keeps to be improved and upgraded from version to version. Some issues mentioned in our paper got improved before the paper was published. The photo-variant approach for dealing with non-metric cameras with unknown orientation parameters was adopted is a good example.

It is true that the paper itself was not submitted to EOS for a review before being sent to PE&RS, although a more detailed Technical Report about the test and evaluation of the software package PhotoModeler was sent to EOS 17 months before the publication of the PE&RS paper. It would be helpful however to suggest to authors that the relevant companie(s) be consulted before publishing a paper.

As a commonly seen application in close range photogrammetry, it is stated clearly in the paper that the tests were made with three types of pre-existing non-metric images taken at different earlier occasions. And it is clear in the paper that the orientation parameters of these cameras were unknown and also were treated as unknown, which is the Inverse Camera method in the Photo-Modeler document, as is also stated in the paper.

As to the issues of camera calibration for non-metric cameras and the issues of Pseudo-Camera concept, they are in fact general issues common to many practical applications and literatures, and are therefore not as the specific negative comments to PhotoModeler. The authors believe that the concepts about these two issues as presented in the paper should be a necessary academic clarification in the world of close range photogrammetry. The authors appreciate the ideas in Mr. Walford’s letter which implies that the stability of the interior orientation parameters of some high-quality non-metric digital cameras are effectively improved compared with the traditional non-metric film cameras. This idea is very probably true, is encouraging to the applications of digital close range photogrammetry, and it is quite worthwhile to set up a project for investigating it.

Gang DENG, Director
Division of International Cooperation
Kunming University of Science and Technology
CHINA

There is a typographical error in the formula for MQNL in Table 2 of my article, “Quantification error versus location error in comparison of categorical maps,” Photogrammetric Engineering & Remote Sensing, 66(8), pp. 1011-1016. The error is an extraneous comma in the formula. The expression should show multiplication between Sj and Rj. It is important that this correction be brought to the attention of PE&RS readers, because it is crucial for readers to understand the mathematical formulation. The following shows exactly how the formula should appear.

Note that

should be

in Table 2, so that the table looks like: