HOG+SVM手势识别.pdf

发布时间：2022-05-30 发布人：admin 分类：说明书资料大小：2.66M 资料格式：pdf 举报版权申诉

daidanzheda-4118604-16359647463718415329.pdf-第1页.png

第1页 / 共69页

daidanzheda-4118604-16359647463718415329.pdf-第2页.png

第2页 / 共69页

daidanzheda-4118604-16359647463718415329.pdf-第3页.png

第3页 / 共69页

daidanzheda-4118604-16359647463718415329.pdf-第4页.png

第4页 / 共69页

daidanzheda-4118604-16359647463718415329.pdf-第5页.png

第5页 / 共69页

daidanzheda-4118604-16359647463718415329.pdf-第6页.png

第6页 / 共69页

daidanzheda-4118604-16359647463718415329.pdf-第7页.png

第7页 / 共69页

daidanzheda-4118604-16359647463718415329.pdf-第8页.png

第8页 / 共69页

文本预览

Computer Vision - June o3, 2oo9 Histogram of Oriented Gradients (HOG) & Support Vector Machines (SVM) HOG: fixed spatial relationships Bernt Schiele TU Darmstadt, Germany http://www.mis.informatik.tu-darmstadt.de/cv/ schiele@cs.tu-darmstadt.de g n i t u p m o C d e t n e m g u A y r o s n e S d n a l a u t p e c r e P

Lecture Overview Recognition (part 1/2) • Object Recognition (Schiele) ‣ Intro - Some basics of digital image processing (2 weeks) ‣ Object Recognition for Identification (1-2 weeks) • Global and local feature approaches ‣ Object Recognition for Categorization (4 weeks) • Object Detection (specific object classes) • Local Features & Interest Points (scale and affine invariant) • Bag-of-Words for Object Categorization • Histograms of Oriented Gradients & Support Vector Machines • Part Representations, Combination with Segmentation Computer Vision - June o3, 2oo9 Bernt Schiele - TU Darmstadt 2

Different Types of Recognition Approaches • Part-Based Object Models consist of: ‣ models / features for object parts ‣ models of spatial layout / toplogy • Object Recognition Methods ‣ Bag of Words Model (BoW) • dense sampling, Interest Points (scale invariant) • Local Features (robustness) ‣ HOG + SVM (today) • HOG = Histogram of Oriented Gradients – global object feature / description • SVM = Support Vector Machines ‣ – discriminant classifier - widely used Implicit Shape Model (ISM) (next week) • local parts & global constellation of parts BoW: no spatial relationships HOG: fixed spatial relationships ISM: flexible spatial relationships Computer Vision - June o3, 2oo9 Bernt Schiele - TU Darmstadt 3

Overview of Today • Histogram of Oriented Gradients (HOG) ‣ global descriptor for object detection ‣ ‣ sliding window approach (slides mostly taken from Dalal’s PhD-defense) • Support Vector Machines (SVM) ‣ general intro ‣ linear SVM Computer Vision - June o3, 2oo9 Bernt Schiele - TU Darmstadt 4

Goals & Applications of HOG • Original Goal: Detect and Localize people in Images and Videos • Applications: Images, films & multi-media analysis ‣ ‣ Pedestrian detection for autonomous cars ‣ Visual surveillance, behavior analysis Computer Vision - June o3, 2oo9 Bernt Schiele - TU Darmstadt 5

Difficulties of People / Object Detection • Some of the 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 for HOG: upright fully visible people Computer Vision - June o3, 2oo9 Bernt Schiele - TU Darmstadt 6

Sliding Window Methods - Overview • Sliding Window Based People Detection: Two Important Questions: 1) which feature vector 2) which classifier ‘slide’ detection window over all positions & scales Focus on building robust feature sets HOG: static and optical flow version Computer Vision - June o3, 2oo9 Bernt Schiele - TU Darmstadt 7 ScanImageExtract Feature VectorClassify Feature VectorNon-MaximaSuppression

Existing Person Detection Methods • Current Approaches ‣ Haar wavelets + SVM: • Papageorgiou & Poggio, 2000; Mohan et al 2000 ‣ Rectangular differential features + adaBoost: ‣ Edge templates + nearest neighbour: ‣ Part-Based Models • Gavrila & Philomen, 1999 • Viola & Jones, 2001 • Felzenszwalb & Huttenlocher, 2000; Ioffe & Forsyth, 1999 Leibe, Seemann & Schiele, 2005; Mikolajczyk et al, 2004 • Orientation histograms ‣ Freeman et al, 1996; Lowe, 1999 (SIFT); Belongie et al, 2002 (Shape contexts) SIFT Computer Vision - June o3, 2oo9 Bernt Schiele - TU Darmstadt 8

分享到：

赞收藏

资料库

HOG+SVM手势识别.pdf

相关推荐

开发技术

热门标签

最新资料