Computer Vision-based Human Motion Capture -a Survey

This technical report is the documentation of a survey within computer vision-based human motion capture. The idea with this report is to present previous work in a structural manner using taxonomies at di erent levels. The report is structured in the following way. The rst chapter gives an introduction to new types of interfaces where one class is known as the "Looking at People" domain. Within this class computer vision-based human motion capture is de ned. In chapter two motion capture is described in a larger context describing the di erent sensors used for motion capture. Chapter three presents previous taxonomies which are commented and a new taxonomy is suggested. Chapter four describes a number of assumptions which are used in computer vision-based human motion capture. The next four chapters each give a detailed description of the four classes present in the taxonomy suggested in chapter three. Finally a conclusion is given in the last chapter. Throughout the descriptions of the taxonomy and its four classes a number of examples are given from the literature. Preface This technical report is the documentation of a survey within computer vision-based human motion capture. The writing of the report was carried out in the winter 98/99 while the reading of papers was done during the summer and fall of 1998. The work presented is the rst step in my Ph.D.study titled: "Multiple Cues for Model-Based Human Motion Capture". A large number of papers have been read to write this report. Approximately 2/3 of them (107) are directly concerned with computer vision-based human motion capture. Detailed summaries of these papers can be found in another concurrently written technical report: Summaries of 107 Computer Vision-Based HumanMotion Capture Papers [108]. Whenever I use 'he' throughout the report it should be read as he/she. Finally I would like to thanks Moritz Stoerring for his help editing the report. Aalborg, Denmark, March 1999 Thomas B. Moeslund ([email protected])