Dear Professor Hartley:
I am a senior student major in photogrammetry at Wuhan University, China. Since I am very interested in computer vision, I am currently taking a research on photogrammetry and computer vision.
Recently I have read your paper "The relationship between photogrammetry and computer vision", SPIE, 1993. I have to admit that something you mentioned in the paper is quiet true. For example, on the PnP problem, "the photogrammetrist refused to even admit the possibility that there could be more than one solution." In fact, I have met this problem in one of my research projects, and I was surprised that, after decades the famous RANSAC paper has been published, in my textbook of photogrammetry, this multi-solution situation is still totally ignored.
After that, my interests on computer vision become deeper and deeper. I have been reading your famous book "Multiple View Geometry in Computer Vision". The compact form of projection matrix and the beauty of projective geometry amazed me. Though the essence of projection matrix in computer vision and collinear equation in photogrammetry are the same, the former reflects more beauty of mathematics.
However, after a little more dig, I found another paper, following your work ten years later, discussing about the relationship of the two fields–"Computer Vision and Photogrammetry–Mutual Questions: Geometry, Statistics and Cognition" by Wolfgang Forstner, a famous photogrammetrist. His paper made me more aware of the relationship, such as that we care more about precision, which is a feature and advantage derived from Geo-surveyings.
In conclusion, both the two papers expressed a hope that there should be more interaction between the two fields. Yet as far as I know, problem still exists. Apparently photogrammetry community has become more and more interested in computer vision approaches and International Society for Photogrammetry and Remote Sensing(ISPRS) have added a Commission III – Photogrammetric Computer Vision and Image Analysis(and the PCV 2010 is about be host in Paris, France). But it seems that little CV researchers pay much attention to this. And ICCV/ECCV have no such commission or workshop.
So, here is my question. It is about two decades after your paper on CV and PH, what’s your comment now on the interaction of the two fields? Is there any "genuine exchange of ideas" as you mentition at the beginning of your paper? If the situation is stil the same, what do you think is the reason, and what can researchers do to improve this?
I am so sorry for being presumptuous to write to you and ask so many (maybe stupid and naive) questions. But I really can’t help it, for I would like to devote my academic career to this interaction and exchange. As an undergraduate student who just steps into research areas, I really want some suggestions from masters like you.
I’m looking forward to your reply. And Thank you very much!
Richard Hartley教授（《Multiple View Geometry》作者）的回信：
Dear Feng Chen,
I apologize for the long delay in replying to your email.
I am surprised that someone actually reads my paper from so long ago.
I think that the situation between photogrammetry and computer vision is significantly improved these days, though the number of people who do interact with both communities is quite small. Of course Wolfgang Foerstner is one of them, highly regarded in both communities. There has of course been a lot of work more recently on all sorts of minimal configurations, such as the work of David Nister, and this is still ongoing. In general, though, there does not seem to be a great hope for future rapprochement of the Comptuer Vision and Photogrammetry communities, partly because Computer Vision has moved on in a way from a major interest in these topics. Perhaps my book is partly to blame for this, since it may have given a perception that the field is very mature. It is, of course, compared to where we were in the early 90s, but there are still people interested in Geometry in Computer Vision. But it is not mainstream any more. The major interest in Computer Vision has returned to topics such as recognition.
In the early 90s, techniques for object recognition (probably the most challenging and important goal in computer vision) have moved from the geometric approach, as put forward by people like Joe Mundy and others. The work of him and others on image invariants, and very geometric-based approaches to recognition led to the major interest in geometry through the 1990s and early 2000s. However, ultimately, that theme of research was played out, and there was little further headway to be made. Current techniques have gone back to less geometric techniques, such as the local feature descriptors, and such, that have proved to be more successful ultimately, though some would still
find them unsatisfactory. The idea of merging recognition using such techniques with geometry is often talked about, but does not seem to be followed much these days.
Anyway, thank you for your email, and I encourage you to keep thinking of such topics
This work, unless otherwise expressly stated, is licensed under a Creative Commons Attribution-No Derivative Works 2.5 China Mainland License.