The focus of this article is that the new modeling software combined with UAV tilt photography has huge advantages. Compared with the traditional modeling method, the efficiency is improved and the cost is reduced. The new method will gradually replace the backward modeling method.
Traditional low-altitude photogrammetry and 3D modeling technology can no longer meet the current requirements for rapid and fine modeling. At present, 3D modeling has become an indispensable and important part of the digitalization process of surveying and mapping. Costs have great advantages in the construction of fine 3D models, the production of large-scale DOMs, and the visual display of 3D models. With the help of oblique photogrammetry technology and Smart3D software, the rapid construction of fine 3D models is realized. The main processes of data acquisition and key data processing technologies are standardized and analyzed, and the accuracy, advantages and disadvantages of using point cloud and image data modeling are compared and discussed. Summarizes the quality requirements and evaluation criteria of relevant data in the process of 3D modeling.
Due to the shortcomings of traditional modeling methods such as low efficiency, high labor intensity, and high production costs, they will gradually be eliminated. Traditional low-altitude photogrammetry technology is widely used in large-area area surveys, safety monitoring, disaster emergency response, environmental protection and many other fields. Through the drone equipped with sensors, high-resolution image data can be obtained quickly, efficiently and conveniently to produce DOM (digital orthophoto) and DEM (digital elevation model). However, due to software and hardware limitations such as sensors and data processing algorithms, the most realistic three-dimensional scene cannot be restored quickly and efficiently.
UAV photogrammetry has the characteristics of flexibility, speed, efficiency, convenience, low cost, and high image resolution, which has greatly promoted the development of tilt photogrammetry technology. Oblique photogrammetry has completely changed the drawbacks of manual modeling, and has greatly accelerated the generation of fine 3D models of large scenes through automated data processing methods. The oblique photogrammetry technology also overturns the limitation of traditional low-altitude photogrammetry that can only obtain data from a vertical angle. The drone is equipped with multiple sensors at the same time to obtain image data from multiple angles, which can more truly and comprehensively reflect the local details of surface objects. And the overall level. At present, Zhongwei Space can obtain rich texture information data through oblique photogrammetry technology , generate dense 3D point clouds and TIN grid models, and combine with automated real-world modeling models to achieve fast, efficient, and low-cost true restoration of 3D scenes . Provided modeling services for many projects .
2 The main process of data acquisition: The following is a description of the process of oblique photography
Tilt photogrammetry mainly consists of three parts: ground flight control system, unmanned aerial vehicle, and control measurement. The flight control part mainly plans and designs the flight route and altitude of the UAV, as well as the flight monitoring and control and data communication of the UAV. The UAV part is mainly composed of an airborne positioning system and a multi-view camera. The control measurement is mainly aerial survey Regional control network design and image control point measurement. Before drone aerial photography, it is necessary to conduct on-site survey of the survey area. First, arrange the image control points reasonably according to the existing GPS control points. The number and location of the image control points are evenly arranged according to the accuracy required by the actual measurement regulations and the size of the survey area. . Secondly, plan the flight route reasonably according to the applied airspace time and scope, and ensure that the course overlap, side overlap, and resolution of the images meet the operational requirements. In the design of the route, a side overlap of 30% and a heading overlap of 66% are generally set. For automatic model generation, side overlap and heading overlap will be more demanding. Once again, the base station must be set up at a known high-precision point, and the UAV must be turned on before the specified time for take-off, and shut down within the specified time after landing. During the measurement, it is necessary to measure the height of the antenna, record the specific time when the base station is turned on and off, and perform the measurement of the image control point. Finally, the drone is assembled and camera parameters are set, and the drone aerial photography is implemented. After the flight, the drone data and base station data are downloaded respectively.
In the process of acquiring image data, there will be unavoidable errors caused by the instrument itself, including the distortion of the camera lens, and the natural influences of the outside world, including weather changes. If the original image is not preprocessed, it will directly affect the accuracy and quality of the post-production data. The subsequent processing of image data is based on digital photogrammetry, multi-view image joint adjustment, computer vision and other related algorithms. The data processing process does not require manual intervention, and has high scalability and efficiency. Using AgisoftPhotoscan, Smart3D capture and other related software, with or without control points, you can achieve multi-view and multi-view 3D reconstruction and restore the most realistic 3D scene. At the same time, the relevant data processing software is used to realize accurate measurement of the height, area, length, volume, etc. of the three-dimensional model. Oblique photogrammetry technology obtains multi-view high-resolution image data, and realizes the most realistic restoration of large scenes and fine three-dimensional scenes. The main flow of data acquisition processing is shown in Figure 1.
3 The key technology of data processing
3. 1 Multi-view image dense matching and air triple solution
Because the images obtained by oblique photogrammetry have a wide range and multiple angles of view, the image field of view between each belt is quite different, and there are often large geometric distortions between oblique stereo images, which increases the difficulty of image matching. The dense matching of multi-view images is the process of finding connection points to construct a network, while eliminating redundant information in the multi-view image data. Image matching algorithms are divided into three categories: gray level matching, feature matching and relationship matching. The commonality of matching is to find points with the same name on the image according to the matching strategy. Based on the feature matching represented by the SIFT algorithm, the matching error is more and the time-consuming is longer. Import and process image data in oblique photogrammetry, and add POS data at the same time to assist the matching of multi-view images. According to POS data, the outer orientation elements of the original image can be roughly obtained, and some mismatched points can be eliminated by rough matching of related algorithms, and then restarted. Exact match. The empty three solution is the process of reconstructing precise topological relations between images. According to the image control points placed on the ground and based on the collinear equation, the beam method area network adjustment is carried out.
3. 2 Realization of multi-node parallel computing
Parallel computing is to decompose a computing task into multiple parallel subtasks and assign them to computing nodes with parallel processing. The processors on each node cooperate with each other to jointly solve the parallel subtasks, thereby accelerating the calculation. Parallel computing system mainly has three important components: parallel machine, parallel algorithm and parallel programming, as shown in Figure 2. The basis of parallel computing is the parallel machine, and the core components of the parallel machine are the processor, memory and interconnection network. The parallel machines are connected in series through the Internet, and the synchronization, sharing and access of image data are realized on the parallel machines. Designing interconnection network topology for specific application types can greatly improve parallel computing capabilities and efficiency. The main design of the parallel algorithm is divided into four steps: task decomposition, communication design, task aggregation and processor mapping. According to the parallel algorithm, the program is compiled in the parallel programming environment and run to obtain the calculation result.
The dense matching of image data and the three-dimensional solution can be implemented on any parallel machine. In the model reconstruction process, the model is divided into several regular tiles of equal size and length. According to the parallel algorithm and program, the serial parallel machine can perform parallel calculation on the divided regular tiles at the same time through the interconnection network. Through the implementation of parallel computing, the speed of calculation and generation of the 3D model is greatly improved, and at the same time, the configuration requirements of the computer hardware for the 3D model are reduced.
3. 3 LOD visualization for GPU
The visualization of 3D models in oblique photogrammetry requires the coordination and cooperation of CPU and GPU. Texture mapping, model rendering, and scene rendering mainly rely on the performance and efficiency of GPU. The GPU has a small cache multi-core architecture and fast and efficient parallel computing capabilities. The data structure that adapts to the GPU must be able to give full play to the GPU's high-speed processing and efficient rendering capabilities to avoid computer hardware data bandwidth conflicts. Model data generated by oblique photogrammetry is processed into blocks and grading, and a quad-tree or octree spatial index model is established for the generated tile data, thereby improving the efficiency of data reading, reducing data I/O operations, and speeding up data Scheduling and drawing. The multi-level-of-detail model (LOD) based on the quad-tree index structure is shown in Figure 3. In the process of 3D model data generation, the LOD of 3D model data is obtained through different simplified scales. Generally, there are at least 5 to 6 layers, and as many as 10 layers.
The next part is data quality analysis and comparison. This part is relatively boring, so I won't repeat it. In summary, facing the large-scene 3D model quickly generated by oblique photogrammetry technology, further development and utilization of the model is still needed. Model singulation and data fusion are the urgent problems facing oblique photogrammetry.