Presentation is loading. Please wait.

Presentation is loading. Please wait.

K.U. K.U. & Leuven & Leuven 2 Computer Vision Labs Prof. Luc Van Gool ETH - Switzerland Un. Leuven – Belgium appr. 15 researchers Tracking Recognition.

Similar presentations


Presentation on theme: "K.U. K.U. & Leuven & Leuven 2 Computer Vision Labs Prof. Luc Van Gool ETH - Switzerland Un. Leuven – Belgium appr. 15 researchers Tracking Recognition."— Presentation transcript:

1 K.U. K.U. & Leuven & Leuven 2 Computer Vision Labs Prof. Luc Van Gool ETH - Switzerland Un. Leuven – Belgium appr. 15 researchers Tracking Recognition Recognition Passive 3D Active 3D Hum.-comp. interact.

2 K.U. K.U. & Leuven & Leuven An overview of some vision trends  scene reconstruction  recognition  tracking

3 K.U. K.U. & Leuven & Leuven Scene reconstruction  3D acquisition with off-the-shelf HW  4D capture: dynamic 3D  Realistic texture synthesis  City modelling  Intuitive visualisation

4 K.U. K.U. & Leuven & Leuven One-shot ShapeCam 3D acquisition

5 K.U. K.U. & Leuven & Leuven 3D acquisition

6 K.U. K.U. & Leuven & Leuven 3D from hand-held camera images

7 K.U. K.U. & Leuven & Leuven... The result generated by ARC3D

8 K.U. K.U. & Leuven & Leuven 4D acquisition 3D snapshots in fast succession

9 K.U. K.U. & Leuven & Leuven Shape-from-silhouettes

10 K.U. K.U. & Leuven & Leuven Outdoor visual hulls

11 K.U. K.U. & Leuven & Leuven Realistic texturing Stochastic & hierarchical texture models Viewpoint/illumination dependent texture Minidome: portable photometric stereo

12 K.U. K.U. & Leuven & Leuven IKT Realistic texture Given examples Synthetic textures

13 K.U. K.U. & Leuven & Leuven AUTOMATIC Realistic texture

14 K.U. K.U. & Leuven & Leuven Recognition

15 K.U. K.U. & Leuven & Leuven Object recognition  Independent of viewpoint  Irrespective of occlusion  In the presence of scene clutter  Under variable illumination  Robust against deformations Latest techniques based on `invariant regions’

16 K.U. K.U. & Leuven & Leuven The ellipses show invariant regions, they cover the same part of the scene The crux is that they were found independently Object recognition

17 K.U. K.U. & Leuven & Leuven Object recognition  Example application: automatic annotation of video data  E.g. finding same object somewhere else in a complete movie Searching for the van in `groundhog day’

18 K.U. K.U. & Leuven & Leuven Automatic retrieval of all scenes with the van based on the example image

19 K.U. K.U. & Leuven & Leuven Object recognition  Next challenge: categorisation  i.e. not recognising particular objects, but rather the class an object belongs to, e.g. a car, a person, etc. This is more difficult, because of the Intra-class variability…

20 K.U. K.U. & Leuven & Leuven Object recognition  Next challenge: categorisation

21 K.U. K.U. & Leuven & Leuven Finding people  Security / surveillance / annotation e.g. pedestrian detector

22 K.U. K.U. & Leuven & Leuven Tracking  Robust blob tracking – anti-drift  Body pose tracking  Detailed hand tracking  Action recognition  Gait analysis

23 K.U. K.U. & Leuven & Leuven Multi-feature tracker

24 K.U. K.U. & Leuven & Leuven HandyMouse project Skin color Detection, Tracking, and Gesture analysis For Minority Report style interaction

25 K.U. K.U. & Leuven & Leuven Marker-less motion capture

26 K.U. K.U. & Leuven & Leuven Spin-offs / start-ups 1. ICOS 2. Eyetronics 3. GeoAutomation 4. eSaturnus 5. Kooaba 6. Procedural


Download ppt "K.U. K.U. & Leuven & Leuven 2 Computer Vision Labs Prof. Luc Van Gool ETH - Switzerland Un. Leuven – Belgium appr. 15 researchers Tracking Recognition."

Similar presentations


Ads by Google