Analyzing Video Sequences of Multiple Humans Tracking, Posture Estimation and Behavior Recognition

Analyzing Video Sequences of Multiple Humans: Tracking, Posture Estimation and Behavior Recognition describes some computer vision-based methods that analyze video sequences of humans. More specifically, methods for tracking multiple humans in a scene, estimating postures of a human body in 3D in real-time, and recognizing a person's behavior (gestures or activities) are discussed. For the tracking algorithm, the authors developed a non-synchronous method that tracks multiple persons by exploiting a Kalman filter that is applied to multiple video sequences. For estimating postures, an algorithm is presented that locates the significant points which determine postures of a human body, in 3D in real-time. Human activities are recognized from a video sequence by the HMM (Hidden Markov Models)-based method that the authors pioneered. The effectiveness of the three methods is shown by experimental results.

1 Introduction.- 2 Tracking multiple persons from multiple camera images.- 2.1 Overview.- 2.2 Preparation.- 2.4 Algorithm for Multiple-Camera Human Tracking System.- 2.5 Implementation.- 2.6 Experiments.- 2.7 Discussion and Conclusions.- Appendix: Image Segmentation using Sequential-image-based Adaptation.- 3 Posture estimation.- 3.1 Introduction.- 3.2 A Heuristic Method for Estimating Postures in 2D.- 3.3 A Heuristic Method for Estimating Postures in 3D.- 3.3.6 Summary.- 3.4 A Non-heuristic Method for Estimating Postures in 3D.- 3.5 Applications to Virtual Environments.- 3.6 Discussion and Conclusion.- 4 Recognizing human behavior using Hidden Markov Models.- 4.1 Background and overview.- 4.2 Hidden Markov Models.- 4.3 Applying HMM to time-sequential images.- 4.4 Experiments.- 4.5 Category-separated vector quantization.- 4.6 Applying Image Database Search.- 4.7 Discussion and Conclusion.- 5 Conclusion and Future Work.