| Summary: | 碩士 === 國立成功大學 === 測量及空間資訊學系碩博士班 === 98 === Image sequences acquired with a Mobile Mapping System (MMS) contain rich information along the trajectory. The coordinates of interested points can be determined by space forward intersection of the images, whose interior and exterior orientation are determined from the navigation and calibration data. Hence the acquisition of spatial data is much efficient. The basic idea to obtain the 3D coordinates of conjugate points from overlap images is to rebuild the geometric relation between images. Obtaining conjugate points of image sequences by image matching is much more efficient than manual measurement. However the factors of scale variations, different field of view, and occlusion may result in incorrect matching. In this research, image matching in object space with virtual surface is proposed to overcome the problems of matching in image space.
Due to the image sequences are acquired during the vehicle moving, the distance variations from objects to cameras cause differences between foreground and background in images. Occlusion problems also cause difficulties of image matching. The process to filter out occluded images is proposed, and the remaining useful images are kept for multiple image matching. In addition, multiple windows matching, in which the target point is located in different positions, is developed to ease the problems caused by complex backgrounds.
The test results of the experiments show the proposed method can deliver correct matching results good to about 80% to 90%. The determined coordinates are about 10 cm in precision, and about 50 cm in accuracy. The quality of point determination is good enough for general applications of GIS data acquisition. Accuracy can be improved, if the positioning and orientation system of the MMS can be improved.
|
|---|
