Structure and camera calibration from active correspondence of lines and intersection points

Many applications, e.g., motion planning, virtual reality, CAD, vehicle navigation, object recognition, photogrammetry, remote sensing, etc., all require a geometrical representation of the three dimensional structure of a scene. In this dissertation we study the problem of determining the 3D structure of a scene given images of it from various views.;We built Biclops a two camera directed vision system. Each camera is mounted on a Pan-Tilt Unit which can be independently controlled. Biclops is built as a tool so that it can be picked up by an industrial robot. This eye in hand system is used to find the structure of the scene. Two robots are mounted on a Robot Transport Unit (RTU) on either end. We use lines in the images and find their intersection points and use them to make the necessary correspondence between the points in different views. Lines and point features and their corresponding entities are used to determine the 3D structure of the scene. The 3D structure is transferred to the other robot space for it to access the objects in the scene.;The accuracy of 3D structure of the scene found is dependent on the accuracy of the various parameters of the underlined equipment. We go a step further to improve the accuracy of the parameters of various equipments used. We developed a new method called ViCKi (virtual close loop kinematic method) to calibrate the industrial robots and the RTU equipment using a laser pointer. We also calibrate the pan tilt units of the Biclops using the laser pointer tool. Various experiments are done to show the accuracy of the calibration methods. This calibrated equipment is used to find the 3D structure of a scene. Various experiments are done to prove the concept that using lines and intersection points are better than using the traditional corner features. The robot uses this 3D structure to pick up the objects. We also used the lines and intersection points to find a track painted on the ground. A robotic mobile platform is used and a vision system is built. The vision system takes pictures of the track and finds its 3D location. The RMP plans it trajectory to follow the track without losing the track out of sight.