Abstract
Real-time understanding of human behavior in video streams is presently one of the most active areas of research in Computer Vision and Artificial Intelligence. Its purpose is to automatically detect, track and describe human activities in a sequence of image frames. Challenges in this topic of research are numerous and sometimes very difficult to work out. Consequently, the progress is very slow and the results are not very satisfactory. This paper aims to survey the methods used in human behavior understanding, showing their strengths and weaknesses. This small "toolbox" of methods and strategies could be very useful to the researcher and the engineer alike.