"Perceiving layout and knowing distances: The integration, relative potency, and contextual use of different information about depth" by James E. Cutting and Peter M. Vishton, enumerates and discusses the nine cues in great detail. The use of 3D depth maps has shown improvement in object segmentation and recognition, so perhaps computer vision has just not attempted the same level of sensor fusion that is necessary for scene comprehension similar to human capability. Cameras operate at 100 megapixels today (similar to rod and cone count in the human eye) and compute capability is rapidly advancing, so it would seem advancement could be achieved in more sophisticated intelligent vision using more cues.
About the same time I was working on the paper you cite, Landy, Maloney, Johnson, & Young were working on what they called "weak fusion" of five cues, attached. [Landy, M. S., Maloney, L. T., Johnston, E. B. & Young, M. J. (1995). Measurement and modeling of depth cue combination: In defense of weak fusion. Vision Research, 35, 389-412.]
Sorry or the hiccup. I haven't followed the area in the last decade, but I would be surprised if computer scientists hadn't worked some of this through.
Thank you for the pointer and response. My group at CU Boulder and here at U. of Alaska is interested in approaches to take with our software and FPGA development and our intent is to pursue the passive methods rather than active depth mappers, so I think this will be a very useful direction that deserves our further study.
Got it and I will share with my student researchers. The fundamental research is quite interesting and informative. Much appreciated.
Hello,
Check these references about integration of visual cues:
*Title Integration of Visual Modules: An Extension of the Marr Paradigm
Authors John Yiannis Aloimonos, David A. Shulman
Publisher Acad. Press, 1989
Original from the University of Michigan
Digitized 16 Nov 2007
ISBN 0120530201, 9780120530205
Length 322 pages
* The MIT Vision Machine (1988)
by T. Poggio , D. Geiger , T. Caw , W. Yang , J. Little , D. Weinshall , H. Biilthoff , A. Hurlbert , E. Gamble , M. Villalba , M. Drwnheller , D. Beymer , W. Giett , N. Larson
Sam,
I hope these references help.
Sincerely this is a very important topic still not resolved.
Greetings,
Homero
Homero - thank you for the pointers to the integration of visual cues. I agree and appreciate the importance - the active RGB depth mappers are great, but I think long term pursuit of passive methods and a better understanding of human vision and the fundamental cues that we process is as you point out, critical to adaptive vision systems that generalize well and are cost effective and power efficient. When I hear about home/elder care robotics, intelligent transportation and augmented reality, my thought is that we still have quite a bit more to learn about interaction and perception common to daily life.
Sam, I agree with you.
As many people have pointed out, many living beings are the best example of integration of visual cues and putting this information into action !.
Even "simple" creatures as fly can do amazing things with their visual system !
Sam,
Long time ago I worked on integrating stereo and shading. I didn't read Cutting's article, it seems interesting, but only few cues from the nine can be implemented. Despite the importance of solving the vision problem, CS researchers gave up working on it. In the 90s it gained a great attention that faded quickly, may be due to the complexity of the problem OR may be due a wrong approach to solving the problem.
I thing separating recognition task from space reconstruction task is the main problem with the current paradigm. WE need a paradigm shift to solve the vision problem.
Nice helpful discussion, I'm really interested in this area and will keep you posted about the research.
- Badges
- Science topic
- Similar topics
- Correction