A study on object location stimation by vanishing points and other shape features for computer vision applications
|關鍵字:||電腦視覺;物體定位;消失點;拋物線;三度空間滑鼠;直昇機降落;自動車定位;computer vision;object location;vanishing point;parabola;3D mouse;helicopter landing;vehicle localization|
The field of computer vision in which visual sensors such as cameras are utilized is highly recognized as having great potential to simulate human eyes. In the field of computer vision, one of the most important problems is object location estimation which means to determine the relative position and orientation between a camera and a set of features such as points detected on an object. Among various kinds of features, the vanishing point is a desired one due to its good properties. In this study, one purpose is to extend the family of features by proving a new theorem about the vanishing point of a parabola. Another purpose is to design some new approaches, using vanishing points and other shape features, to solve the problem of object location estimation for various computer vision applications. Four approaches have been proposed, which deal with the following applications: design of three-dimensional mice, location estimation of helicopter landing sites, and localization of autonomous vehicles. In the following, for the derived theorem and each of the proposed approaches, the employed methodology, the experimental results, and some concluding remarks are described. The vanishing point of a parabola is a new shape feature found in this study. In the new theorem about the vanishing point of a parabola, six properties of this new shape feature are proved. Traditionally utilized features include points, lines, vanishing points of lines, and so on. The discovery of the vanishing point of a parabola broadens the family of features and promotes the ability of utilizing various kinds of environmental information for computer vision applications. More and more three-dimensional input devices are required for human-computer interaction in the world of virtual reality. In this study, an inside-out vision-based three-dimensional mouse is proposed. The basic way to manipulate this type of device is to hold a camera to view a square mark in front of it. This is a new idea since all other vision-based three-dimensional mice are outside-in. Automatic landing systems can alleviate heavy burdens on aircraft pilots. In this study, a new approach is proposed to estimate the location of a helicopter landing site. All other existing vision-based methods for aircraft landing are based on motion analysis of a sequence of images. The proposed method instead achieves aircraft location estimation by using monocular images. The principle is to utilize information contained in an international identification mark on the helicopter landing site. One desirable way for vehicle navigation is to utilize information existing in environments. Among various kinds of indoor information sources, ceilings have the good properties of being less occluded and having uniform patterns. The third approach proposed in this study achieves vehicle localization by utilizing ceiling information. The location and the lateral position of the vehicle are derived by using the parallel lines on the ceiling. Finally, while the former method is for indoor vehicle navigation following a straight line, it is also desired to conduct vehicle localization in an outdoor road turning session. And this goal is achieved by the fourth proposed approach. The idea is to view a curve as a parabola and to let the vehicle follow the curve in the turning process. Experiments have also been conducted to prove the correctness of the derived theorem and the feasibility of all the proposed approaches.
|Appears in Collections:||Thesis|