Abstract
Under the background mature applications of artificial intelligence techniques, the development of image, vision and three-dimensional processing techniques promote the process of environment digitalization of smart cities, and the digitalization of plants has attracted much attention. The research contents of tasks of the project are in eight aspects: vegetation data acquisition, vegetation visual information analysis, 3D point cloud reconstruction, high-detail reconstruction and interactive reconstruction of vegetation models based on data and mechanism, urban scene analysis and scene reconstruction, optimal representation of 3D vegetation, fast realistic rendering of vegetation scenes, and forestry applications.
The project has published 71 papers, including 22papers on IEEE/ACM Transactions journals, 21other SCI journal papers and 4 top conference papers. Among all the published papers of the project work, 21 were marked as the first acknowledgements.
The project has 20 representative papers, reflecting typical achievements and major contributions in four aspects. (1) In the aspect of visual analysis and point cloud reconstruction, we proposed methods on sparse spectral unmixing, fine structure segmentation of natural images, image processing accelerated by bilateral filtering, and joint computation of image segmentation and depth estimation. (2) In the aspect of 3D plant model reconstruction, we proposed methods on synthetic modeling of the tree trunks and twigs, plant growth simulation and interaction with the environment, reconstruction of blooming flowers, deformation capture and modeling, realistic procedural modeling of plants from multiple view images and interactive plant surface reconstruction from incomplete point clouds. (3) In terms of fast rendering and optimal representation, we proposed methods on direct construction of plant continuous LOD models, hardware instancing rendering based on the GPU, surface point sampling and surface re-meshing. (4) In the aspect of scene reconstruction and forestry application, we proposed methods on automatic constraint detection for 2D scene layout regularization, inverse procedural modeling of facade layouts, and precise measurement of stem diameter by simulating the path of diameter tape.
Also a monograph was published for the project and 25 patents have been applied or authorized in China and abroad. Achievements of the project has won the second award for progress in science and technology by Ministry of Education of China, and the first award for progress in science and technology by Department of Education of Shaanxi Province. Supported were a total of 18 doctoral students, 6 master students, and 5 postdoctoral scholars.
The project has developed new methods and new technologies for image processing, computer vision and 3D geometry processing to solve the technical problems related to data acquisition, data processing, data reconstruction and presentation of vegetation objects in the process of urban digitization. Results of the project are of great values in the fields of urban digitization, ecological assessment and spatial layout planning.
Keywords: segmentation and depth estimation; synthetic modeling; procedural modeling; continuous LOD models; hardware instancing rendering