Title: Towards Detecting, Recognizing, Understanding and Interpreting of Human Activities

Abstract: This talk will introduce visual applications which are related to human-machine interaction. Person following with a mobile robot is concerned with one of the critical challenges of video based tracking namely the ability to selectively focus attention of the system on the object of interest against a varying background. The person extraction and tracking takes place by analyzing color image streams and utilizing the stereovision based depth images of the scene. A linear Kalman filter is utilized to provide the estimates of position and to perform a smooth tracking. A real-time face detection algorithm will be presented in context of pointing-arm posture recognition as well as visual tracker (re)initialization without constraining the users. First the algorithm locates the person's face and afterwards some arm-postures are recognized. The start as well as the end of the pointing command has been signaled by suitable static arm-postures. Selected dynamic commands have been recognized on the basis of HMM. The use of a particle filter combined with color, depth information and shape features as en efficient and effective way to deal with tracking of a head on the basis of image stream coming from a mobile stereovision camera will be outlined next. Experimental results that were obtained on long image sequences indicate the feasibility of such an approach to perform tracking of a head undergoing complex changes of shape and appearance. Finally, initial research findings that are connected with use of a particle filter and labeling techniques as en efficient and effective way to deal with inter-occlusions of the tracked face and hands during tracking will be overviewed. The problem how to apply omnidirectional images as well as histogram matching/Markov localization techniques in order to obtain information about position of mobile agent in real-time will be discussed as well.