| Along with the social informatization, the demands to the real-time and low-cost high-resolution images are increasing rapidly. Unmanned Aerial Vehicle Low Altitude Remote Sensing System (UAVRS) is known as a system with the unmanned airships as its vehicle platforms. UAVRS combines many high technologies such as modern electronics, computer studies, communications, controlling, remote sensing and telemetering with unmanned airship. It possesses the abilities of instant observation to the earth, immediate processing of RS data, as well as acquiring high temporal and spatial resolution images at low cost, which is a very significant and indispensable complementary to traditional aerial and satellite remote sensing. Therefore UAVRS is widely used in Quick Disaster Response, city construction monitoring, precision agriculture, environmental monitoring, investigation of soil utility status and map revision, etc.With the support of the project"Key Technologies and Application Examples of Geological Survey of Remote Sensing based on the High Resolution Images Obtained by Low Altitude Unmanned Airship"of Ministry of Land and Resources, this paper introduced the preprocessing of low altitude remote sensing images acquired by UAVRS and realized the preprocess of image and a key technology of automatic mosaicking of panoramic images using photogrammetry method, image matching, as a foundation for the further geological analysis.The preprocessing is an essential phase for the automatic mosaicking of panoramic images. For the reason that the digital cameras in UAVRS are not specially made for Photogrammetry, the elements of interior orientation are unknown, hence the images, which could be affected by the camera distortion and not suitable for complete automatic image mosaicking, need preprocessing. The errors of cameras are mainly caused by the distortion of optical lenses, called optical lens distortion, will lower both the quality of images and accuracy of processing. Therefore, the calibration of optical lens distortion is absolutely necessary in image preprocessing.The DPMatrix system developed by Wuhan University was used to correct the camera distortion and get the elements of interior orientation as well as distortion factors by using the line mathematical equations. After resampling with bilinear difference method, the image was obtained through distortion correction, thus provides the preparation for the next step.Image matching, which is based on the image feature points extracting, is a key technology for the automatic mosaicking of panoramic images. At present, the often-used feature extraction operators include Moravec operator, Forstner operator, Harris operator, Trajkovic operator and the SIFT operator. After comparing feature points extraction algorithm's speed, accuracy and overall flexibility, the Harris operator was found to be of the most stable, robust and the highest accuracy. In this study, the high precision Forstner operator and faster feature extraction Harris operator were combined, and then the Harris improved algorithm was gotten. The Low Altitude Images has high spatial resolution, and in order to improve the running speed and make a Uniform distribution of feature points, not only the information-rich regions were focused on. The original image (2848pixel×4288pixel) was divided into 220 small units (13pixel×19pixel), and for each unit, the feature points were extracted with Harris operator. Eventually, a total of 220 points were Extracted, achieving sub-pixel accuracy.On the base of extracting image feature points, it's very important to match corresponding points accurately and fast. Corresponding points were matched in the right image by grayscale matching algorithm under the feature points extracted in the left image. The method of pyramid hierarchical searching was used to enhance operational efficiency and reduce operation time because of the high resolution of the low-altitude unmanned airship's image. After low-pass filter in the original image, the rough positions of corresponding points were found by Rough correlation, and the results were used as predictive values .Then high-frequency information were added step by step, with the precise correlation in the smaller and smaller areas. For the top layer of the pyramid, a two-way matching was used to ensure the matching accuracy, then the matchings in the middle layers were decided by the correlation coefficient method, at last the matching was refined by the least square in the bottom layer of the pyramid to improve the accuracy of image matching. In the matching process, the features points passed from bottom to top while the matched corresponding points passed in the opposite direction. Eventually, a total of 178 corresponding points with their own coordinates were obtained, providing the solid foundation and Necessary prerequisite for the realization of automatic mosaicking of panoramic images... |
PDF Full Text Request