Tele-Existence System Based On Real-Time Camera Calibration

It has long been a desire has been there, people hope to sit at home or office can control a robot in the remote environment, feeling real time sensory information which is being feedback by robot through the sensors on board the robot, so that people can have real sensory of the remote environment. On the other hand a teleoperated robot system is useful for people in various situations such as maintenance of plants or power stations,operation of construction work,care of elderly or handicapped people and supply of aid in the case of an emergency or disaster. Especially in some hostile environments such as nuclear,high temperature,deep space, people want to remote operation to Completed tasks by controlling a salve robot, to avoid unnecessary damage.With the rapid development of robot technology and network communication technology as well as the emergence of precision measuring instruments,the appear of tele-existence meet the needs of the people.It can make people to feel the real time sensory of being present or exist in a different place other than the place they are really exist at the same time.Realization of a visual display with a sensation of presence is one of the most important elements of tele-existence. Current tele-existence technical research can be divided into two cases:one through slave robots in remote environment to get the real time sensory in real remote environment, another is through the scene synthesised by computer to get real presence in the virtual reality.In the 21st century, the tele-existence technology become more and more maturation, people can operate a remote robot to complete a variety of complex work, and can get high real feeling of real time sensory in the remote environment. But in the research of tele-existence technology, the information of human head movements is captured by the sensor, which need to install expensive sensors on the helmet display and the tracking accuracy isn't limited. This paper presents a vision-based tele-existence system. We use camera calibration algorithm to calculating video-based helmet camera parameters,which can solve the head motion tracking problem under the fixed viewpoint.Using camera calibration to capture the head movement information, we must solve two problems:real-time and tracking accuracy, so We don't only need to ensure the accuracy of the calibration,but also need to make sure the time delay of the algorithm enough small. Now the accuracy of existing camera calibration algorithm calibration is high, but in general the time delay is relatively large, does not meet the requirements of tele-existence system. So this paper proposes a GPU-based fast fixed-viewpoint calibration algorithm to achieve fast calibration on the low configuration computer, and time delay be control in a very small range. The basic method is:using SIFT feature points to organize and express the offline scene, and using KLT algorithm to track feature points on the online video sequence. And at the same time matching SIFT feature points for camera initialization and the elimination of accumulated error,so that the calibration algorithm has a fast calculate rate. As GPU-based feature extraction techniques were improved, making the time of feature extraction and registration greatly reduced, and makes the time delay control of system in a very small range.