Abstract:
Human-robot interaction is a very fundamental need for assistive and service robots. In order to interact with human beings, detection and tracking of people is a basic requirement. A great deal of work has been done in this area and various sensors have been utilized to reach the solution. Generally laser range finder and cameras have been the most common devices used for human detection but still there is a huge problem of human tracking in the presence of multiple human beings and at various illumination levels.
This research work presents a new approach of human detection and targeted person tracking by a mobile robot. The major contribution is the development of a novel Motion Detection based Tracking scheme which utilizes the differential images of three iterative frames of live video stream. Then the comparison is made with three other conventional techniques; cam shift, LK optical flow and particle filter. The robot is equipped with stereo camera and sonars. Human detection is done using Haar cascade classifier, followed by the implementation of a tracking method on the selected target person. Comparisons of these techniques are based on testing in different environmental conditions. Afterwards, Kalman filter based prediction and correction is done to accurately localize the target. Moreover stereo correspondence is done by template matching followed by linear triangulation, to get the 3D location of the target.
Thus, this research work presents the implementation of various techniques that make the robot to select a target person in the environment and then track and follow that target. Results of all these methods are presented and their evaluation is done in detail.
