1. 1 of 25 1 of 22 Blind-Spot Experiment Draw an image similar to that below on a piece of paper (the dot and cross are about 6 inches apart) Close your right eye and focus on the cross with your left eye Hold the image about 20 inches away from your face and move it slowly towards you The dot should disappear!

2. 2 of 25 2 of 22

3. Mobile Robotics: 7. Vision 2 Dr. Brian Mac Namee (www.comp.dit.ie/bmacnamee)www.comp.dit.ie/bmacnamee

4. 4 of 25 4 of 22 Acknowledgments These notes are based (heavily) on those provided by the authors to accompany “Introduction to Autonomous Mobile Robots” by Roland Siegwart and Illah R. Nourbakhsh More information about the book is available at: http://autonomousmobilerobots.epfl.ch/ http://autonomousmobilerobots.epfl.ch/ The book can be bought at: The MIT Press and Amazon.com The MIT PressAmazon.com

5. 5 of 25 5 of 22 Today’s Lecture Today we will have a quick tour of some of the uses of vision sensors in robotics today, including: –Colour models –Object recognition –Face recognition –Stereo vision –Object tracking We will give a brief overview of all of this as computer vision is far too massive a subject to cover in our course

6. 6 of 25 6 of 22 RGB Colour Model Think of R, G, B as color orthobasis (0,1,0) – pure green (0,0,1) – pure blue (1,0,0) pure red (1,1,1) - white (0,0,0) - black (hidden)

7. 7 of 25 7 of 22 HSV Colour Model More robust against illumination changes Still must confront noise, specularity etc.

8. 8 of 25 8 of 22 Object Detection Suppose we want to detect an object (e.g., a colored ball) in the field of view We simply need to identify the pixels of some desired colour in the image…right? Image coordinates (pixels) u v IOIO width height

9. 9 of 25 9 of 22 It’s Not That Easy! Occluded light source Specular highlights Mixed pixels Complex surface geometry (self- shadowing) Noise!

10. 10 of 25 10 of 22 Evolution Robotics’ ViPR System Evolution Robotics ViPR (visual pattern recognition) technology provides a reliable and robust vision solution to object recognition The technique is based on extracting salient features from an image Salient features are artefacts such as edges, corners etc The description of an object is a set of up to a thousand salient features, the textures of the pixels around them and their relationships to each other

11. 11 of 25 11 of 22 Evolution Robotics’ ViPR System (cont…) Object recognition then involves first finding all of the features in a new image These features are matched against those of all of the models in a database If many features in a new image are the same as those in a database model, the model is a good candidate match Further accuracy is obtained by comparing the relative positions of the matched features in the image and the model The model in the database with the best match score is then recognised

12. 12 of 25 12 of 22 ViPR Example

13. 13 of 25 13 of 22 ViPR Pros & Cons The advantages of the ViPR system include: –Invariance to rotation and affine transformation –Invariance to changes in scale –Invariance to lighting changes –Invariance to occlusions –Reliable recognition However, while the ViPR can be used to recognise symbols and 3D objects it cannot be used to recognise deformable 3D objects such as faces For more information on the technologies behind ViPR have a look at: “Core Technologies for Service Robotics”, N. Karlsson, M. E. Munich, L. Goncalves,Core Technologies for Service Robotics J. Ostrowski, E. Di Bernardo & P. Pirjanian “Distinctive Image Features from Scale-Invariant Keypoints”, D. LoweDistinctive Image Features from Scale-Invariant Keypoints

14. 14 of 25 14 of 22 ViPR Demo

15. 15 of 25 15 of 22 Vision Through Colour Tracking Often colour alone can be used to perform vision tasks We use flood fill techniques like those in Photoshop Particularly useful in controlled environments

16. 16 of 25 16 of 22 Face Tracking Using Colour Alone Image AcquisitionRGB to HSV ConversionSkin Colour Binary ImageImage ClosingSegmentationSelection by Size

17. 17 of 25 17 of 22 Stereo Vision Stereo vision is used to determine distance through differences in images taken from two cameras positioned slightly apart –Just like our eyes!

18. 18 of 25 18 of 22 Stereo Vision Idealized camera geometry for stereo vision

19. 19 of 25 19 of 22 Stereo Vision A point visible from both cameras produces a conjugate pair –Conjugate pairs lie on epipolar line (parallel to the x- axis for the arrangement in the figure above) From a conjugate pair distance can be estimated

20. 20 of 25 20 of 22 Stereo Vision Calculation of Depth –The key problem in stereo is now how do we solve the correspondence problem? Gray-Level Matching –Match gray-level wave forms on corresponding epipolar lines –“Brightness” = image irradiance I(x,y) –Zero Crossing of Laplacian of Gaussian is a widely used approach for identifying feature in the left and right image

21. 21 of 25 21 of 22 Stereo Vision Example Depth image (bright = close, dark = far) Confidence image (bright = high confidence) Vertical edge filtered left and right images Original left and right images

22. 22 of 25 22 of 22 Summary Today we looked at some of the vision techniques commonly used in robotics We have barely scratched the surface, but hopefully you have gained some appreciation of the difficulties involved

23. 23 of 25 23 of 22 Questions? ?