Model Reconstruction Algorithms For Massive Point Cloud From Terrestrial Laser Scanner

Compared to traditional acquisition methods for spatial information, terrestrial laser scanning technology can obtain high-precision and high-density3D coordinate information of target actively and integrally with the superiority of non-contact, all weather and high efficiency. So, it provides a more richer data support for model reconstruction, especially for the attractive objects such as urban building facades and indoor targets. Therefore. it plays a unique role in three-dimensional reconstruction. However, the theory of point cloud processing for object-oriented modeling lags be-hind the rapid development of hardware and can not meet the sharply increasing requirements of efficiency for various engineering applications. It has become a bottleneck and restricts the further application and development of terrestrial laser scanning technology.Therefore,in this thesis.aiming at an efficient and common solution of massive point cloud processing for the reliable reconstruction of typical urban indoor and out-door targets using terrestrial laser scanning data.deep and systematical research has been conducted on the key issues including massive data management.data segmenta-tion.classification and model reconstruction.The main contents and achievements could be summed as follows:Effective organization and management of massive point cloud data is the pri-mary key issue that need to be addressed in the efficient processing of point cloud data. At first. the basic principles of distance measurement. angle measurement and positioning of terrestrial laser scanning are introduced in this thesis.as well as the analysis of point cloud characteristics and its usual storage and interchange format. On this basis. a new organization method of massive scattered point cloud data using random sampling is presented.followed by the corresponding processing strategies for the support of efficient interaction. neighbor searching and dynamic updating. The implemented experiments show that our method obviously outperforms other existing ones and is suitable for the efficient processing of massive point cloud.The plane segmentation is a hotspot and also one of the difficulties in the point cloud data processing,which plays an important role in point cloud registration, object reconstruction and recognition, machine vision, etc. Based on the introduction of point cloud segmentation and classification of current methods, the advantages and shortcomings of commonly used mehtods are analyzed. For point cloud data of large-scale complex scenes.traditional ways of segmentation become inefficient or even inapplicable.So.given the principles of efficient segmentation, a novel hybrid approach based on octree structure is presented to improve the efficiency and reliabil-ity simultaneously of plane segmentation.The comparative experiments verify our method is automatic and robust in the segmentation process.In addition.on the basis of data feature analysis. a new pre-filtering method is put forward for the further improvement of efficiency.which is combined with data management to achieve effi-cient segmentation of massive point cloud.Based on the segmentation of point cloud. we analyze the characteristics of plane segmentation for typical targets in terrestrial scanning scene.followed by the discus-sion of a novel approach for the construction of distinctive feature vector.In order to get reliable data classification results. the SVM has been studied systematically and an original classification method which synthesizes rule-set based means and the SVM is proposed. Classification experiments are conducted on different scenes.and the results verify its reliability.Furthermore.the main factors of classification accuracy are discussed and summarized.In order to implement the efficient and reliable reconstruction of complex scenes. based on the results of data segmentation and classification, a hierarchical and intelli-gent hybrid strategy using the selection of the optimal model has been established for the automatic fitting and reconstruction of urban targets of interest.For the improve-ment of efficiency and reliability of the reconstruction process. the process of model fitting optimization is discussed in detail.as well as the reconstruction flowchart.The experiments reveal that the proposed approach could support high-precision and relia-ble reconstruction. At last. we give out the further research directions of model reconstruction.On the basis of independently developed hardware of panoramic camera. we conduct research on the registration of point cloud and panoramic image from the vehicle-borne laser scanning system.mainly including the panoramic camera's internal and external calibration. After the establishment of point-to-image mapping model, a dual model is proposed for urban rapid measurement.with the integration of point cloud and panoramic image.The presented approach is evaluated using field check points and the results demonstrate that the accuracy of measurement could meet the needs of the urban information collection.And this could lay a solid foundation for the further integrated processing of point cloud and image.At last,an integrated and mapping-oriented software is developed for the massive point cloud processing.In this part,we introduce the overall module designation and the critical data structures.along with some typical applications in urban surveying and mapping. And it could be used as a solid platform for the further engineering applications of terrestrial laser scanning technology.