Line, a stability characteristic of images, is one of the important study objects of computer vision. Line detection and matching plays a very important role in many tasks of computer vision, such as image registration, 3D reconstruction, object recognition and video understanding, etc. Although a variety of works have been done in the field, there are still a lot of challenging problems. This thesis is focus on line detection, line matching and 3D reconstruction. The main work is as follows: A new approach based on inner product of gradients for line detection is presented. Firstly, the inner product energy of image points is introduced, which not only can enhance line edges but also restrain image noise and non-line edges. Then, an edge detection method with enhancing lines is proposed by replacing the gradient argument with the inner product energy. Lastly, the inner product of gradients is used to extract lines form the detected edges. Experiments show that, in addition to a better performance of noise resistance, the approach performs better than the phase-grouping algorithm widely used in literatures in detecting low contrast lines. The algorithms based on gray correlation and gradient descriptor for ling matching are introduced and implemented. The experimental results show that, in the match number and the correct rate of matches, the algorithm based on gradient descriptor is superior to that based on gray correlation. Based on the line matching, for the completeness of lines originally detected, a line linking approach is proposed. It uses both line descriptor and position information to obtain a more complete line. Experimental results confirm the effectiveness of this method. Lastly, the results of line matching are used for 3D reconstruction. Experiments show that, for man-made scenes, the precision of 3D reconstruction can be improved by the match lines.
修改评论