Research of Axel Pinz

Research Projects of Axel Pinz

On this page you find summaries of what I currently consider my most important contributions.
A complete list of past and current research projects can be obtained from my cv.

Since Nov. 1999, I have been building up a new group on real time measurement and tracking. This has also been the main area of my research over the past years. See tracking@emt for a detailed description.

Recent developments in our tracking technology and algorithms are nicely demonstrated by the following videos:

1. High-speed tracking by a custom CMOS camera. The camera has a USB2 interface and tracking rates of up to 2.5 kHZ are achieved by direct subwindow access. video (5MB)
2. Mobile tracking gear used in front of a building. Tracking and 3D graphics augmentation of the facade. video (15MB). Check also the paper (pdf, 3.5MB)


In an invited talk at VMV'01 (Vision, Modeling and Visualization), I gave an overview on "Consistent Visual Information Processing" which refers to most of the work on active object recognition and on tracking described below.
paper (pdf, 650k)
Visuals of the talk (ppt, 3.57MB)
In the talk, 4 videos were presented, which can also be downloaded from this page:
Active vision laboratory video (MPEG, 20MB)
Active object recognition video (MPEG, 35MB)
Tracking of an interaction panel, sequence 1 video (M2V, 28MB)
Tracking of an interaction panel, sequence 2video (M2V, 28MB)
The tracking videos show that real-time tracking can be achieved by application of spatial constraints (video 1) and that multiple hypotheses can be further reduced by temporal filtering (video 2).

Image Understanding - Active Fusion - Object Recognition

Starting in 1985-88 with my dissertation on VES, a Vision Expert System, I have been continuously interested in high level vision. I have been teaching courses in Image Understanding since 1986, and also written a book on that subject ("Bildverstehen", Springer, 1994).

A major basic research project employing a research group of up to 5 persons has been carried out under my supervision in 1994-1999. Within this project, we developed the concept of Active Fusion and applied it to the area of Active Object Recognition. An active observer can take many different images of a scene. Information extracted from these images has to be evaluated, selected and combined with respect to a goal of the system. We have compared probabilistic, possibilistic, and fuzzy approaches and developed a method which can be used to control the active vision system to select the next (most promising) view. The method has been applied and evaluated on appearance based object recognition, where our results demonstrate the ability to plan disambiguating views in cases of ambiguity (several different objects looking similar from a certain viewpoint).
The system has won the best demonstration award at the British Machine Vision Conference in 1998. For more details we provide a paper (pdf, 750kB) and two videos:
Active vision laboratory video (MPEG, 20MB)
Active object recognition video (MPEG, 35MB)

My current and future research interests in this area are towards generic active object recognition.


Real-time Tracking: Augmented Reality - Hybrid Tracking - Mobile Collaborative AR

Research in this area deals with problems of precise spatial registration of a device (eg see-through AR helmet) in six degrees of freedom (position and orientation, 6DoF) and in real-time (>20 Hz). We have developed a new, hybrid tracking system consisting of a commercially available magnetic tracker for coarse but fast and reliable estimation, and vision-based tracking for real-time refinement. The system is currently operated in a multiuser office environment called "Studierstube" and has been developed in collaboration with the Institute of Computer Graphics at Vienna University of Technology
(-> visit Studierstube/ICG Vienna).

Within the European TMR Project VIRGO, our experience in fusion and in localisation has been successfully applied to the task of mobile robot localisation.

We are a scientific partner of the VRVis competence center for Virtual Reality and Visualization. The center is funded within the KPlus initiative of the Austrian government, and mostly located in Vienna. Our contribution to VRVis is in basic research in the area of Tracking. We will contribute to all kinds of optical tracking (outside-in and inside-out; tracking of artificial and of natural landmarks; use of artificial IR lighting technology; gesture and person tracking), to the use of new camera technology (CMOS) for optical tracking,  as well as to the development of a suite of new inertial sensors as a complementary sensing system.

In Nov. 2000, we (Dieter Schmalstieg at Vienna University of Technology and Axel Pinz at Graz University of Technology) have started on a new research project funded by the Austrian Science Fund (FWF), entitled "MCAR - Mobile Collaborative Augmented Reality". In this project we investigate on tracking requirements for fully mobile collaborative AR. Each user will be equipped with a wearable AR kit using "Studierstube" technology. Magnetic tracking will be replaced by a new combination of sensors including vision and inertial tracking.


Application Projects
  • Quantification of posterior capsule opacification (PCO): After cataract surgery of the human eye, PCO constitutes the most relevant post-operative complication. A quantitative image analysis package called AQUA, for the grading and segmentation of different textures in retro-illuminated PCO images is currently being developed.
  • Real-time quality grading of apples: A novel sensory concept and image analysis algorithms for real-time (5 apples per second per production line) inspection of apples have been developed. Quality grading includes assesment of size, shape and color as well as the detection of defects (hailmarks, russet, rotten spots).
  • Active inspection system for steel quality assessment: Steel specimen are taken during the processing of steel (rolling, heating, etc.). These specimen are polished, etched, and microscopically inspected. Steel quality is reflected by scaled texture properties, which has been successfully modeled by Gabor filters and by scale space. online paper (.pdf 1MB)
  • Mapping the Human Retina: information from several imaging modalities of a scanning laser ophthalmoscope (SLO) is fused to obtain a retinal map for improved diagnosis and treatment of age-related macular degeneration. online paper (pdf 1.2 MB)
  • Detection and classification of trees in aerial images: Starting with my PhD thesis in this area, I have contributed to the field of high resolution remote sensing imagery for forestry, especially to the delineation of individual trees and to tree species classification.