Object Tracking for Calibration of Distributed Sensors in Intelligent Space

In recent years, the research field on smart environments, which are spaces with multiple embedded and networked sensors and actuators, has been expanding (Cook & Das, 2004). The main purpose for the introduction of smart environments is to support humans in both physical and informative ways (Johanson et al., 2002), (Mynatt et al., 2004), (Mizoguchi et al., 1999). In smart environments, the distributed sensor nodes observe the spaces, extract useful information from the obtained data and the actuators including mobile robots provide various services to users. Moreover, robots in the space can get necessary information from the smart environment and operate without restrictions due to the capability of on-board sensors and computers. In fact, mobile robot control is easier in smart environments since the global positions can be directly measured by using distributed sensors and Simultaneous Localization And Mapping (SLAM) problem (Durrant-Whyte & Bailey, 2006), where the robot tries to simultaneously estimate its own location and build a map of the environment, can be completely avoided (Lee & Hashimoto, 2003). However, one of the major problems in developing smart environments is calibration of the sensors. Calibration is needed for proper calculation from the local (sensor's) coordinate system to the world (smart environment's) coordinate system and it is usually done by using calibration objects that are objects with known positions, shapes and so on. For example, researches on camera calibration based on geometrical features including 3D points (Tsai, R. Y., 1987), 2D points (Sturm & Maybank, 1999), (Zhang, 2000), lines (Zhang, 2004), circles (Wu et al., 2004), and spheres (Agrawal & Davis, 2003) are being pursued actively. Several researchers extended such a single camera calibration algorithm to a multiple-camera calibration algorithm. An extension of a planar point pattern based calibration algorithm to the multi-camera systems with an arbitrary number of cameras is presented in (Ueshiba & Tomita, 2003). The algorithm is based on factorization of homography matrices between the model and image planes, which is expressed as a composition of a camera projection matrix and a plane parameter matrix. If a calibration object is put in three or more places, the relative positions and orientations between cameras as well as the intrinsic camera parameters are determined. Another work also utilized a known 2D calibration object for stereo camera calibration (Malm & Heyden, 2001). The technique uses both stereo camera constraints and single camera constraints, and therefore noise-robust calibration is realized. In the case of smart environments, it takes a great deal of time and efforts to calibrate the sensors since multiple sensors are distributed in the space. Although the researches