Submitted by: Submitted by abc3217630
Views: 232
Words: 2028
Pages: 9
Category: Business and Industry
Date Submitted: 06/16/2013 01:36 AM
Finding People in Images and Videos
Navneet DALAL
GRAVIR, INRIA Rhône-Alpes Thesis Advisors Cordelia SCHMID et Bill TRIGGS
17 July, 2006 Institut National Polytechnique de Grenoble
Goals & Applications
Goal: Detect and localise people in images and videos Applications:
Images, films & multi-media analysis Pedestrian detection for smart cars Visual surveillance, behavior analysis
2
Difficulties
Wide variety of articulated poses Variable appearance and clothing Complex backgrounds Unconstrained illumination Occlusions, different scales Videos sequences involves motion of the subject, the camera and the objects in the background Main assumption: upright fully visible people
3
Talk Outline
Overview of detection methodology Static images
Feature sets Object localisation Extension to other object classes
Videos
Motion features Optical flow estimation
Part based person detection Conclusions and perspectives
4
Overview of Methodology
Detection Phase
Scale-space pyramid Scan image(s) at all scales and locations Extract features over windows Run linear SVM classifier on all locations Fuse multiple detections in 3-D position & scale space Object detections with bounding boxes
5
`
Detection window
Focus on building robust feature sets (static & motion)
Finding People in Images
N. Dalal and B . Trig g s . His tograms of Oriented G radients for Human Detection. C VPR , 2005
6
Existing Person Detectors/Feature Sets
Current Approaches
+1 -1
Haar wavelets + SVM: • Papageorgiou & Poggio, 2000; Mohan et al 2000 +1 -1 Rectangular differential features + adaBoost: • Viola & Jones, 2001 Edge templates + nearest neighbour: • Gavrila & Philomen, 1999 Model based methods • Felzenszwalb & Huttenlocher, 2000; Ioffe & Forsyth, 1999 Other works • Leibe et al, 2005; Mikolajczyk et al, 2004
Orientation histograms
Freeman et al, 1996; Lowe, 1999 (SIFT); Belongie et al, 2002 (Shape contexts)
7
Static Feature Extraction...