Research On Close-Range Photogrammetry Based On Multi-baseline Normal Digital Images

The digital images taken by normal digital cameras not only have advantages of low cost, high-resolution, feasible for data acquisition, high speed to transmit and process, but also make it possible to acquire images quickly in the engineering fields, what's more, can lower the requirement for equipments and technology of close-range photogrammetry. Recently, enormous researches have been made on the three-dimension data extraction of close-range objects based on normal digital cameras. However most researches follow the scheme of taking use of single image pair to match points first and then solving their three-dimension coordinates by self-calibration bundle adjustment or Direct Linear Transformation (DLT). These methods have some shortcomings such as high demands for the number and the distribution of control points, low reliability of point-matching and so on. Hence, this paper researched on digital close-range photogrammetry based on multi-baseline normal digital images. This paper has contents and results as follows:(1) The shortcomings of recent metric application are analyzed, and a sequential parameters estimation method based on preference calibration modal selection and perspective transformation is proposed to metricilize normal digital camera, which includes the determination of distortion calibration coefficients and inner orientation factors of normal digital camera. Experimental results prove that the method can resolve the coefficients by less control points, and avoid the relativity between orientation factors and distortion calibration coefficients. The metric accuracy of normal digital camera will enhance obviously after the metricilization.(2) For the remained distortion error of normal digital camera after the metricilization, and multi-baseline images' features of short baselines and multi-overlap, this paper researches on the relative orientation based on approximate epipolar lines. This method allows for efficiency and accuracy simultaneously, and can match enough relative orientation points based on few conjugate points to ensure the accuracy.(3) Stereo matching method for multi-baseline images and the stereo restraint between spatial points are studied. This method can match points sequentially, and improve matching reliability.(4) Based on normal digital images, the accuracy and reliability of "space resection—multi-images space intersection" and multi-images bundle adjustment under the condition of different observations and different control conditions are discussed. Experimental results prove that bundle adjustment of metricilized normal images is the best algorithm for normal images.(5) For the close-range photogrammetry adjustment of free-network that lacks of reference value, the adjustment of free-network algorithm is proposed, which takes unknown as observation and the reliability analysis is made on this algorithm based on Helmert-WF variance component estimation. Experimental results prove that the result of this algorithm is unbiased, and the initial values have less affect on it, and it can improve the resolving speed and the reliability of the adjustment.