CSTalks - Object detection and tracking - 25th May


Published on

Object detection is a fundamental step in most of the video analysis applications. There are many research challenges involved in automatic object detection, depending on different scenarios. The most prevalent application of object detection is in the field of multimedia surveillance. In this talk we will discuss the common problems in the object detection in a surveillance video. Further, we will discuss the Gaussian Mixture Model (GMM) based object detection method. While object detection is the basic step of video analysis, higher level semantic interpretation of the scene requires trajectory information. Most of the suspicious event detection methods use tracking as the basic building block. In the second part of the talk, we will discuss particle filter based method of object tracking. To summarize, the aim of the talk is two-fold: (1) Discuss common problems in object detection and tracking (2) Hands on experience of how to use classical methods of GMM and particle filtering in problem solving.

Published in: Technology, Business
  • Be the first to comment

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

CSTalks - Object detection and tracking - 25th May

  1. 1. Tutorial 7Object Detection and Tracking<br />Mukesh Saini<br />1<br />
  2. 2. Layout<br />Object detection<br />Challenges<br />GMM based method<br />Tracking<br />Challenges<br />Particle filter based tracking<br />2<br />
  3. 3. Object Detection<br />Goal: To detect the regions of the image that are semantically important to us:<br />People<br />Vehicle<br />Buildings<br />Application<br />Crowd management<br />Traffic management<br />Video compression, video surveillance, vision-based control, human-computer interfaces, medical imaging, augmented reality, and robotics…<br />3<br />
  4. 4. Object Detection in Images<br />Subjectively defined<br />Generally template based<br />Mainly done by image segmentation<br />4<br />
  5. 5. Object Detection in Videos<br />Relatively Moving – Object<br />Relatively Static - Background<br />5<br />The goal here is to differentiate the moving object from background!<br />
  6. 6. Surveillance Video<br />Static camera<br />Background relatively static<br />Subtract the background image from current image<br />Ideally this will leave the moving objects<br />This is not an ideal world…<br />6<br />
  7. 7. Feature Based<br />Objects are modeled in terms of features<br />Features are chosen to  handle changes in illumination, size and orientation<br />Shape based – Very hard <br />Color based – Low cost but not accurate<br />7<br />
  8. 8. Template Based<br />Example template are given<br />Object detection becomes matching features<br />Image subtraction, correlation<br />8<br />
  9. 9. Motion Based<br />Model background<br />Subtract from the current image<br />Left are moving objects <br />Remember! This is not a real world…<br />9<br />
  10. 10. Problems in Modeling Background<br />Acquisition noise<br />Illumination variation<br />Clutter<br />New object introduced into background<br />Object may not move continuously<br />10<br />
  11. 11. Outline of Object Detection<br />Determine the background and foreground pixels<br />Draw contours around foreground pixels<br />Use heuristics to merge these contours<br />11<br />
  12. 12. Ideal World<br />Single value modeling of background<br />Anything different is foreground<br />12<br />
  13. 13. Static Background<br />Each pixel resulted from a particular surface under particular lightening<br />Single Gaussian is enough ( )<br />If<br />Pixel belongs to background, else foreground<br />13<br />Background<br />Foreground<br />Foreground<br />
  14. 14. Whenever a pixel matches the background Gaussian, update the background model i.e.<br />If<br />Then<br />Standard deviation updated accordingly<br />14<br />Illumination Variation<br />
  15. 15. Clutter<br />Think of tree leaves…<br />Multiple surfaces, still part of background<br />Gaussian Mixture Model<br />Update each Gaussian after matching<br />15<br />
  16. 16. Static Object Introduced<br />Think of flower pot…<br />Background model should adapt to this change<br />Use Gaussian for new surface as well<br />Few extra Gaussians for the foreground<br />16<br />
  17. 17. Measuring Persistence<br />Modeled as prior weight w<br />More persistent Gaussians belong to background<br />If a new pixel does not match to any exiting Gaussians, least persistent Gaussian is replaced with a new Gaussian with:<br />And standard variation <br /> = a large value<br />17<br />
  18. 18. Background Selection<br />A background Gaussian will have<br />More persistence – high w<br />Less variation – low <br />Sort Gaussians wrt<br />Pick top k Gaussians as background such that<br />If pixel belongs to one of these, it’s a background pixel<br />18<br />
  19. 19. Adaptive Background Model<br />Every pixel is modeled as mixture of Gaussians<br />More persistent Gaussians belong to background and others to foreground<br />The Gaussians are updated after each frame<br />19<br />
  20. 20. Connecting the Dots<br />The output of background modeling is a binary image<br />Dilation/Erosion can further reduce noise<br />Contour drawing<br />Bounding boxes <br />20<br />
  21. 21. Revisit the problems<br />Problems<br />Slow moving background – clutter<br />New object introduced into background<br />Illumination variation<br />Object may not move continuously<br />21<br />
  22. 22. Thank You<br />Q & A<br />22<br />