Recent Developments in Human Motion Analysis Liang Wang, Weiming Hu, Tieniu Tan Chinese Academy of Sciences, Beijing, People’s Republic of China 2002 Pattern.

Recent Developments in Human Motion Analysis

Liang Wang, Weiming Hu, Tieniu Tan

Chinese Academy of Sciences, Beijing,

People’s Republic of China

2002 Pattern Recognition

Outline

• Potential application• Detection• Tracking• Behavior analysis• Future researches

Potential Applications

• Visual surveillance• Tracking and recognition techniques of face

and gait• Advanced user interface

• Control and command by speech, gestures, body poses, facial expressions, etc.

• Motion-based diagnosis and identification• Medical diagnosis, sports, orthopedic patient

s, choreography

Motion Detection

• Human detection aims at segmenting regions corresponding to people from the rest of an image.

• Motion segmentation• Background subtraction• Statistical methods• Temporal differencing• Optical flow

Motion Detection

• Object classification• The purpose of moving object classification is to pr

ecisely extract the region corresponding to people from all moving blobs obtained by the motion segmentation methods.

• Shape-based• NN classifier

• Motion-based• Periodic property• Residual flow

Human Tracking

• Useful mathematical tools• Kalman filter• Condensation algorithm• Dynamic Bayesian network

• Different classification• Hand, face, leg, whole body• Single-view, multiple-view, omni-directional view• 2-D, 3-D• Indoors, outdoors• Single human, multiple human, human groups• Moving, stationary• Monocular, stereo

Human Tracking

• Model-based• Stick figure (fig.)• 2-D contour (fig.)• Volumetric models (fig.)

• Region-based (fig.)• Active-contour-based (fig.)• Feature-based

Recognition and Description of Human Activities• Behavior understanding is to analyze an

d recognize human motion patterns, and to produce high-level description of actions and interactions.

• General techniques• Dynamic time warping (DTW)• Hidden Markov models (HMMs)• Neural network (NN)

Recognition and Description of Human Activities• Action recognition

• Template matching• State-space approaches

• Semantic description

Further Researches

1) Segmentation

2) Occlusion handling

3) 3-D modeling and tracking

4) Use of multiple cameras

5) Action understanding

6) Performance evaluation