[11.4,5] Object Tracking: A Survey

ACM Computing Surveys (CSUR) 2006
Alper Yilmaz, Omar Javed, 그리고 이름도 익숙한 Mubarak Shah

목차

1 Introduciton
2 Object representation
3 Feature selection for tracking
 - Color
  . RGB dimensions are highly correlated.
  . Luv and Lab are perceptually uniform color spaeces, while HSV is an approximately uniform color space. However, these color spaces are sensitive to noise.
 - Edges, Optical flow, Texture
4 Object detection
 4.1 Point detectors
  - SIFT and KLT
 4.2 Background subtraction
 4.3 Segmentation
  4.3.1 Mean-shift clustering
   - find clusters in joint spatial+color space [l,u,v,x,y]
  4.3.2 Graph-cuts
  4.3.3 Active contours
 4.4 Supervised learning
5 Object tracking
 5.1 Point tracking
  5.1.1 Deterministic methods for correspondence
  5.1.2 Statistical methods for correspondence
   5.1.2.1 Single object state estimation
    - Kalmal filters
    - Particle filters
   5.1.2.2 Multiobject data association and state estimation
    - Joint probability data association filter
    - Multiple hypothesis tracking
 5.2 Kernel tracking
 5.3 Silhouette tracking
6 Related issues
7 Future directions
8 Concluding remarks

 

감상

양이 매우 많은데 (40페이지) 일단 맛보기로 훑어봄.
particle filters, MHT, mean-shift, graph-cut에 대한 맛을 봄 (맛없음).