Research On Key Technologies Of DOM Generation By Aerial Video

In recent years, low-altitude aerial photogrammetry based on unmanned aerial vehicles (UAV for abbreviation) has becoming juxtaposed with satellite remote sensing and traditional aerial photogrammetry. Its main objective is to meet the modern society for the timely mapping and fine mapping applications. Despite the advantages of timely and accurate, the UAV usually need to wait for landing to get the image data, and cannot be able to meet the timely requirements for emergency mapping. To solve the problem, the video camera is integrated in the sensor system, and using wireless transmission technology for real-time transmission. However, based on the small size, small field angle, low resolution, the video image cannot directly reflect the overall profile of the survey area, limiting its potential applications. Therefore, the study of using video image to create orthophoto has an urgent application requirements and research value.Using aerial video to create orthophoto is mainly related to two aspects:First, how the dynamic video is to converted to the static image; Second, how to use the frame image to create orthophoto without initial position information. For the first question, its need to study the key frame extraction technology and the overlap between the frames must meet the mosaic requirements. The second problem can be divided into two sub-problems. First, the position and attitude of the frame image must be restoration. Secondly, stitching the ortho rectified images, to create a geometric consistent, color smoothly orthophoto. Therefore, the main work and contributions of this thesis are as follows:(1) First, the geometric transformation model of the2D and the3D to2D are summarized and its accuracy difference are analyzed. Secondly, the method of parameter estimation is given. Combined with the advantage, geometric accuracy, and application requirements, the thesis is designed two kinds processes for orthophoto production, one is based on image side, the other is based on object side.(2) The method and process of static geometric calibration for video camera are studied. First, based on the analysis of the distortion low for the video camera, through the comparison of a classic distortion model of camera and the Brown model, a new distortion model is proposed to take into account higher-order terms and cross terms, so that the overall static video camera calibration accuracy is less than0.5pixel. Secondly, the video camera calibration process is to be derived, analyzing the results of the proposed distortion model and verifying its validity.(3) Based on the principle of curve fitting, an adaptive method for key frame extraction is proposed. By analyzing the overlap variation of UAV video, a two-step algorithm for key frame extraction is proposed:the learning phase and the extraction phase. The objective of the learning phase is calculated variation of overlap between frames on the current flight and terrain conditions. The extraction phase sampling to calculate the degree of overlap within a specified range, and the curve of the overlap is been fitted by the initial value of the learning phase provided. Then calculated the frame index which overlap meets the requirements accordance with the fitting results.(4) A method of using image process for blurred image restoration is proposed, which caused by image motion. First, a detailed summary of the research progress about the kernel estimate and the image deconvolution is presented. Secondly, based on information entropy, signal-to-noise ratio, and information content, the thesis established the reference and no-reference evaluation system for the restored image. The Experiments show that the restoration method is effective.(5) A method for the structure of the frame image and global optimization is proposed based on the image side. Through the analysis of the motion model and its error propagation law, and according to the characteristics of frame images without POS information, a method is proposed by the properties of the tie point, which the key frame gradually added into the image sequence and the structure and motion has been recovered. Through the principle of least squares to make the accumulated errors assigned to the image sequence which has been added. So that all points of the project error is minimized. Finally, the experimental results demonstrate the effectiveness of the method.(6) A method of CPU and GPU co-processing for aerial video stitching is proposed. In order to meet the needs of timely mapping for aerial video stitching, and based on the research progress of the parallel processing for the remote sensing data, we propose a method using GPU parallel techniques to accelerate the key steps. According to the characteristics of CPU mode and GPU mode, we designed a process flow to make the two types of processors for joint operation.(7) Through a specific application example, the stitching flow of aerial video is discussed and some related data products are evaluated, reflecting the capabilities and the feasibility to use the aerial video for orthophoto production by comparing the results of two models. The aim of this paper is to resolve the key problems in the process of generation orthophoto by using aerial video. The paper combined with much key technologies such as video camera geometric calibration, key frame extraction, image motion recovery, image structure from motion construction, GPU parallel computing, and using key frame image complete orthophoto production based on image side and on object side, expanding the scope of application of aerial video.