CC BY
42
GEOGRAPHICAL SCIENCES
BUILDING OBJECT-BASED VIRTUAL MODELS BASED UPON TERRAIN LASER SCANNING AND UAV DATA Rylskiy I.A.1, Markova O.I.2, Eremchenko E.N.3, Panin A.N.4 (Russian Federation)
Email: [email protected]
'Rylskiy Ilya Arkadievitch — PhD in Geography, Senior Researcher; 2Markova Olga Ivanovna — PhD in Geography, Senior Researcher; 3Eremchenko Eugeniy Nikolaevitch - Researcher, WORLD DATA SYSTEM, GEOGRAPHICAL FACULTY; 4Panin Alexander Nikolaevich — PhD in Geography, Senior Researcher, Associate Professor, RESEARCH LABORATORY OF INTEGRATED MAPPING GEOGRAPHICAL FACULTY, LOMONOSOV MOSCOW STATE UNIVERSITY, MOSCOW
Abstract: virtual modeling is a rapidly developingtechnology that use spatial information. Any kind of spatial data is potentially suitable for creating virtual models. The simplest and cheapest to obtain are virtual models with the territory represented as a single surface, not divided into separate objects. Such models are often done according to aerial photography from UAVs, but this is extremely inconvenient, inefficient and is a dead end approach — especially when modeling large areas.
In this paper, we consider a different approach to creating virtual models based on aerial photography from UAVs, laser scanning of the terrain, and an object-oriented approach. Buildings and structures are considered as separate objects associated with the database and hyperlinks. The accuracy of the reference corresponds to the accuracy of topographic plans on a scale of': 500. In accordance with the foregoing, we have produced several models of urban centers in the Russian Federation. Keywords: LIDAR, spatial data, geoinformatics, UAV, virtual modeling.
СОЗДАНИЕ ВИРТУАЛЬНЫХ МОДЕЛЕЙ С ИСПОЛЬЗОВАНИЕМ ДАННЫХ НАЗЕМНОГО ЛАЗЕРНОГО СКАНИРОВАНИЯ И СЪЕМОК С БПЛА Рыльский И.А.1, Маркова О.И.2, Еремченко Е.Н.3, Панин А.Н.4 (Российская Федерация)
'Рыльский Илья Аркадьевич — кандидат географических наук, старший научный сотрудник; Маркова Ольга Ивановна — кандидат географических наук, старший научный сотрудник; 3Еремченко Евгений Николаевич — научный сотрудник, Региональный центр Мировой системы данных, географический факультет; 4Панин Александр Николаевич — кандидат географических наук, cтарший научный сотрудник, доцент, Научно-исследовательская лаборатория комплексного картографирования, географический факультет, Московский государственный университет им. М.В. Ломоносова, г. Москва
Аннотация: виртуальные модели — быстро развивающееся направление в использовании пространственной информации. Для создания виртуальных моделей потенциально пригодны любые виды пространственных данных. Наиболее простыми и дешевыми в получении являются виртуальные модели с представлением территории в виде единой поверхности, не разделенной на отдельные объекты. Подобные модели часто делают по данным аэрофотосъемки с БПЛА, но это крайне неудобно, неэффективно и является тупиковым подходом при моделировании крупных территорий. В данной работе рассмотрен иной подход к созданию виртуальных моделей на базе данных аэрофотосъемки с БПЛА, лазерного сканирования местности и объектно-ориентированного подхода. Здания и сооружения рассматриваются как отдельные объекты, связанные с базой данных и гиперссылками. Точность привязки соответствует точностям топопланов масштаба ': 500. В соответствии с вышеизложенным нами было изготовлено несколько моделей городских центров на территории РФ.
Ключевые слова: лазерное сканирование, пространственные данные, геоинформатика, беспилотный летательный аппарат, виртуальная модель.
UDC 004.67:910.27(075.8) DOI: 10.24411/2542-0798-2020-17103
1. INTRODUCTION
Production of virtual models becomes more and more popular in nowadays. The increased computing power of desktops and mobile devices allows you to use virtual models as a standalone product. It is also possible to use them as a kind of element of augmented reality or as an additional element of complex hardware decision support systems (for example, automated workstations of rescue services)
At the same time, the requirements for completeness and quality of drawing the details of models are continuously increasing, which requires the use of more and more sophisticated methods of modeling and texturing objects (when using manually applied texture makers and manually created models). This leads to an extraordinary appreciation of any virtual models created with the participation of professional model-makers, etc. In the case of virtual models replicated in millions of copies (for example, models used in computer games or popular web services), this approach is economically viable and acceptable. For models with a lower potential number of users, such costs of labor, money and time are unacceptable.
Traditionally low-budget virtual models (during last 20 years) over relatively large areas (more than 100 hectares) were created using GIS data. Usually they are represented by a digital elevation model, which is textured by satellite images. Vector data is represented by linear and area objects (vertically parallelepipeds, fences, walls, lakes, "plates", etc.) elongated vertically, and three-dimensional inscriptions. Sometimes in such low-detail models several (up to 50-100) truly complex three-dimensional objects are inserted.
All of these models use the principle of layer-by-layer data storage (in raster layers, shapes, coatings). In addition to visual asceticism, such models are basically still a set of GIS data, which is visualized in one form or another using some third-party software (for example, Erdas Virtual GIS, SpaceEyes, ArcScene, etc.), without which all this data cannot be viewed in the form of a virtual model. The mentioned software is usually not cheap and cannot be used by a large number of users, including at the same time (Fig. 1)
Fig. 1. Example of a virtual model with 3D elongated buildings and GIS data
At present, such models are of little satisfaction to users both in terms of the quality of the "picture" and in terms of functionality. They are not very high-tech in manufacturing and require, first of all, the availability of spatial data already processed to the level of ready-made GIS-layers.
2. KEY TASKS
As we see, the development of methods for creating virtual models that are traditional for geoinformatics is not very promising, and in the near future, it does not seem to lead to dramatic changes in the field of creating virtual models. However, the use of GIS-layers of raster and vector data to create virtual models (hereinafter - VM) is not a dogma. In fact, the majority of users primarily (unlike map and GIS users) expect from virtual models REALISTIC ENVIRONMENT both in terms of the visual series and in the interaction (for example, the inability to fly through walls, inertia when flying over terrain, the effects of real world like flowing water, etc.). GIS layers (first of all, it concerns vector data) are the quintessence of abstraction, when real objects are reduced to two-dimensional geometry and a database. Historically, when VMs first became possible for ordinary users to create on personal computers (mid 1990s), either map data (raster or digitized
76
vector) or ready-made GIS layers were available for work. Using data from aerial photography or full-fledged three-dimensional models were not even discussed.
For virtual models of significant (hundreds and thousands of square kilometers) spatial coverage, such approaches remain relevant now. However, production of VM of average size (tens of square kilometers) or small (several square kilometers or less) there is a need for much more detailed data. At the same time, in 2019 there are several methods for collecting spatial data, which (in various forms) can be used to create VMs, bypassing the phase of two-dimensional vector GIS data during processing. At the same time, they can in principle be available to the average user for purchase and / or independent receipt. Of these methods should be noted following [1, p. 148]:
- airborne laser scanning and aerial photography data (ALS and aerial imagery) [4, p. 179]
- terrestrial laser scanning (TLS)
- aerial photography (nadir and oblique) captured using the UAV.
In this paper, we focused on creation of a virtual territory model that meets the following requirements:
- the model must be produced using terrestrial laser scanning data and aerial photography from the UAV without the using other materials of remote sensing data and GIS (with the exception of address information);
- to work out the methodology for modeling three-dimensional objects without using two-dimensional vector GIS layers;
- to work out the method of texturing three-dimensional objects using not conditional, but photo-realistic textures;
- create a virtual model in the form of an independent application.
- object-oriented approach of data storage.
3. METHODS OF ACQUIRING SPATIAL DATA
Airborne laser scanning data is more or less similar to the data of terrestrial laser scanning (though it is less dense) and with aerial photography data from the UAV (but may also have less detail). However, with the appropriate parameters of the surveying, they may be very similar in their characteristics to each other. In addition, laser scanning and simultaneous aerial imagery acquisition with UAVs are gradually being introduced into practice. Due to this similarity, we will discuss separately the methods of terrestrial laser scanning (TLS) and aerial imagery with UAVs.
3.1. TERRESTRIAL LASER SCANNING.
The method appeared at the beginning of the 21st century and is essentially a further development of non-reflecting total stations. The principle of operation is quite simple - short laser pulses are used for measuring the distance (non-reflecting method), the horizontal and vertical angles of emission of the beam are also measured. The resulting reflection is recorded in the scanner's own coordinate system, while the reflected signal has a measured value of the amplitude of the reflected signal and can be normalized to the distance, thus obtaining the reflectance of the object from which it is reflected. Oscillating the laser beam in the vertical and horizontal planes allows to survey the terrain in a certain radius from the device.
Modern terrestrial laser scanners have a speed of up to 1,200,000 points per second (Riegl VZ2000i) and a range of up to 6000 m (Riegl VZ6000), while the accuracy of measuring the coordinates of individual points is 5-15 mm at distances of 500-6000 m, with an average distance between points about 5-15 mm. This allows to use ground-based scanning systems to obtain unprecedentedly accurate data on the shape of the surfaces of objects within the field of view of the instrument. In addition, usually ground-based laser scanning is carried out in parallel with the ground-based photography, subsequently these data are combined. Terrestrial laser scanning is very convenient when working on relatively small objects - up to 10 km2. At the same time (depending on the device) only 1-2 people are enough for performing surveys; there is no need to hire a separate aircraft or vehicle, to receive special permits. The cost of equipment these days is also not fantastic.
Object modeling using TLS data requires different methods and software. In case of simple form of objects, it is possible to use the digitization mode in 2D with the subsequent assignment of the height attribute. In case of more complex form of the object, it is necessary to use professional solutions for creating 3D graphics (for example, 3D Studio MAX), CAD (AutoCAD, Microstation) or specialized software (for example, Phidias, Polyworks).
The core advantage of terrestrial laser scanning method for virtual modeling is that it allows you to arbitrarily detail the drawing of captured objects without changing the tool and the result depends mainly on the authors desire and effort. At the same time, the time spent for surveying (in case of using modern equipment) depends little on the requirements for the detail of the final model (in most cases, the level of detail is redundant).
3.2. UAV AIRBORNE SURVEYS.
Perhaps this is one of the most famous innovations in the field of spatial data collection; It is very well known to both professionals and ordinary people, so we will not describe it in detail. Note that the use of compact systems based on multikopter UAVs, equipped with cameras with a frame size of 20 or more
77
megapixels and a phase GNSS receiver for fixing the coordinates of the centers of photographs, allows you to quickly (1-4 days) obtain aerial data in areas of 10 km2 or more by a brigade of 1-2 people.
However, for the purpose of creating a virtual model, the requirements for aerial surveys are somewhat different. Nadir shooting is primarily used to create a digital model of the visible surface (in city conditions, these are surfaces of vegetation, buildings, and relief). On a nadir survey, vertical objects and wall textures are quite poorly visible, although the roofs and horizontal surfaces are clearly visible. To obtain wall textures, shooting at an angle is necessary, that is, perspective shooting. Or oblique shooting.
It was established experimentally that the most versatile solutions in the field of perspective shooting are systems equipped with 5 cameras — one per nadir (with a focal length F providing coverage of about 60 angular degrees) and four oblique cameras installed perpendicular to each other by azimuth and at an angle about 35 degrees to vertical. In this case, it is possible to observe an approximate equality of pixel sizes on the nadir and oblique images, to eliminate the possible "dead zones" of the stereoscopic photographs and to ensure that all objects in a given area are covered with images from all sides.
Traditionally, such systems exist for manned systems (for example, solutions from IGI or LEICA). Since 2015, at exhibitions, individual manufacturers have presented similar solutions (but more compact, with worse characteristics) for UAVs. However, in all cases, these systems are quite expensive and weighty, because they require a corresponding large and expensive UAV.
However, the results similar to those described above can be obtained using a UAV equipped with a single camera. In this case, not one, but three flights are required. In the first of them, the camera is installed in the nadir and the flight is at the optimum height from the point of view of photogrammetry (let's call it H). In the second fly, the height is reduced to 0.7 N, while the camera is set, say, at an angle of 35 degrees to the vertical. Flights are made according to the standard scheme of parallel flights with overlapping (30%) in directions, for example, "north-south". In the third flight, the height and overlap of the flight is set similarly to the second flight, but the axes of the flights are laid in the "west-east" directions. Despite the increase in flight time, this allows the use of much cheaper and compact drones, which in most cases is critical when creating virtual models of small coverage (Fig. 2).
Fig. 2. Example of oblique (perspective) image of model made from UAV
The presence of hardware-measured information about the coordinates of the centers of projections of photographs is very valuable. It allows to significantly speed up the process of aerial phototriangulation and significantly improve the accuracy of the solution. With the joint orientation of nadir and oblique photographs, such information is extremely necessary. At the same time, the accuracy of such measurements should be at a level of 20 cm in plan and in height. In this work we used L1 GNSS receivers while implementing a singlecamera survey.
Laser scanning from the ground has a number of features inaccessible to the photogrammetric method. So, it shows wires, poles, hanging and inclined structures (all this turns into noise and artifacts on photogrammetry data). At the same time, terrestrial scanning is optimal for scanning vertical surfaces. At the same time, airborne imagery made with a UAV is optimal for shooting low-angle (roof) and horizontal surfaces (relief,
asphalt, etc.). Merging these data allows you to eliminate disadvantages of each of the methods by filling in the "dead zones" and additionally displaying new objects.
4. BUILDING THE VIRTUAL MODEL
The described above approaches to the information processing for creating a virtual model will be demonstrated using following example: creation of model of the central part of Vladivostok. This model was created in 2018-2019, the survey was performed using the following types of equipment:
- terrestrial laser scanner Riegl LMS Q620 (shooting distance up to 1000 m), accuracy - 10 mm, scanning speed - 24 000 points per second;
- DJI Phantom 4 Pro UAV with one 20 Mpix camera and an onboard GNSS receiver (L1, phase, improved version of a serial product done by the authors of the article);
- Trimble 5700 GNSS receivers (used for measuring ground control points and geolocation of the laser scan positions data of the terrestrial laser scanner).
The survey was carried out in the central, most mountainous part of Vladivostok (areas near Tigrovaya, Naberezhnaya streets). Along the streets with complex or historical facades, as well as in areas of complex relief (waterfront of Vladivostok), terrestrial laser scanning was performed in increments of up to 0.008 angular degrees. Parallel to this, nadir and oblique aerial photographs were taken using a UAV and RGB camera. The resulting datasets were first processed separately.
4.1. PRELIMINARY DATA PROCESSING
There were 46 ground control points marked in advance (crosses on solid objects, identifiable on photographs and on ground scanning data). Measurements were performed in the local coordinate system. The RMS error of measurements of the ground control points within the local coordinate system was 28 mm in plan and 38 mm in height, which is enough to create materials of 1:500 scale with a relief cross section of 0.5 m (average elevation error less than 15 cm) . To ensure maximum data integrity during adjustment, both the photogrammetric block and laser scanning data used the same control points.
Terrestrial laser scanning data (up to 50 scan positions) were adjusted and geo-referenced using Riegl RiScan Pro software. The results of the processing were clouds of laser reflection points from visible surfaces and the relief of streets and adjacent objects (Fig. 3).
Fig. 3. Example of terrestrial laser scanning data in Vladivostok
Photogrammetric processing of airborne survey data was performed using AgiSoft Photoscan software. The results of aerial phototriangulation allowed us to obtain the following types of data:
- orthophotomosaics (using nadir images);
- textures of the house walls and structures (in the form of textures for OBJ files);
- clouds of points (similar, but not the same as laser points) whose coordinates were obtained from photogrammetric data.
The resolution of the photographs was about 3.5 cm for nadir shots and 4.5 cm for oblique shots. The RMS error on control points was about 9 cm.
4.2. BUILDING SURFACES OF VISIBLE OBJECTS
Combining heterogeneous data of two point clouds (laser and photogrammetric) made it possible to obtain the most complete data on the walls (difficult target for UAV survey but perfect for terrestrial laser scanner )
and roofs (not visible on the terrestrial laser scanning, but perfect for UAV survey). According to the results of the combined point cloud, raster models of surfaces (relief, walls, roofs) were created. The surfaces were presented in a regular (raster) form (pixel size of 15 cm). The data format is ArcInfo GRID. Further, we will call this surface DSM - digital surface model.
Normally, in cheap models, the workflow ends on a stage of forming DSM. Later it is mapped by orthomosaic data, forming "environment". But the resulting model, consisting of millions or even billions of faces, makes great load on the computer during visualizing. And slso, this approach has several problems: very large memory usage (in gigabytes) even for simple and small models; difficulties with importing / exporting to the major part of CAD and GIS softaware (they simply do not support such a number of faces; in the model there is no division for anything: houses, relief, cars, treetops - all of this is merged into one surface model; it is impossible to attach any attribute load on objects due to the absence of objects; degradation of clear spatial forms and edges into smoothed, "swollen" forms.
Such a model, which does not have clearly traced structural lines, along which surfaces are bent, will be referred to as a non-structural three-dimensional model. Such problems make the final result practically unacceptable for any use other than viewing (Fig. 4). To eliminate these shortcomings, it is necessary to actually digitize the necessary spatial edges of objects with the subsequent creation of a model with a much smaller (hundreds and thousands of times) number of faces [5, p. 13].
Fig. 4. Distortion of building forms on a model without structural lines (left) and the correct representation of the same
elements in a structural model
We modeled three fundamentally different classes of objects in different ways:
1. Relief and hydrography. This is modeled by the method of classification of three-dimensional points with the subsequent thinning of the obtained triangulation model [7, p. 937] to the level of 50000-60000 faces per 1 km2 with edge lengths up to 30-50 m. The procedure is largely automated. Used software TerraSolid. Modeled by segments of several tens of hectares.
2. Vegetation was modeled as separate pieces of surfaces ("groups of trees") in the form of a triangulated surface with edge lengths up to 4-6 m. Modeled by segments from 100 m2 to several tens of hectares
3. Buildings and artificial structures. Modeled manually. For this was used DSM. DSM was used for drawing edges of three-dimensional elements of buildings and structures. Modeling of buildings and structures was performed with fixing of the main architectural elements - edges of the structure, forms of roofs, groups of balconies, porches, extensions [3, p. 340]. Estimated accuracy of contour drawing is 20 cm (enough for 1: 500). The estimated height accuracy of the surface model obtained is no worse than 12 cm in height (taking into account the error of the control points themselves).
The model in which the three-dimensional data described above is presented (including the surfaces in which the edges of objects are drawn) will be called the STRUCTURAL THREE-DIMENSIONAL MODEL.
To optimize the efforts to create three-dimensional objects, the following approach was applied. Most of the buildings are relatively simple geometric shapes, developed vertically without overhanging parts (with the exception of roof overhangs and church domes). Therefore, to adequately draw the edges of the modeled object as a whole, it is sufficient to digitize these edges in the top projection ("in plan") and then assign 3D coordinates to each of the vertices of the digitized lines using a digital surface model (DMS)
To produce structural 3D model, we have created a set of GIS-tools based on ArcView. Separate tools were also developed to create models of buildings with overhanging roofs and domes, allowing them to be modeled with parameters (size of the overhang, diameter of the dome), and using DSM in raster data format (roof dimensions, height of the bottom and top of the dome, etc.).
Fig. 5. Groups of buildings represented in a structural 3D model with photo-realistic textures
After digitizing in such a "2.5D" mode, an OBJ object is constructed for each structure (building). Thus, instead of an inseparable surface, we get separated three-dimensional objects, that allows us to go to the object-oriented approach when creating a virtual model [6, p.40]. The resulting 3D objects (in the OBJ format) are added to the previously created photo triangulation project Agisoft Photoscan for automatic generation of wall and roof textures. This allows us to apply textures to the faces of each object, ensuring geometric correctness and photorealism of the models (Fig. 5).
Using DSM, created with an accuracy of 1: 500 and higher, allows you to create 3D objects with a similar accuracy, apply real textures to their walls and bring these objects into a virtual model. The ability to measure in 3D allows you to measure the area and distance on the facades, the length of wires between buildings, and so on.
4.3. CREATION OF STAND-ALONE VIRTUAL MODEL
The obtained three-dimensional objects with traced edges (mainly buildings), relief models and groups of trees with the corresponding textures were used to create the virtual model itself. The model was created using the UNITY environment. UNITY - a cross-platform development environment for various kinds of computer simulations. Unity allows you to create applications that run on more than 20 different operating systems, including personal computers, game consoles, mobile devices, Internet applications and other
The main advantage of Unity is a visual interface of environment construction, cross-platform support and a modular component system. The disadvantages are difficulties when working with multicomponent schemes and difficulties in connecting external libraries. At the same time, Unity is used by both large developers and independent studios.
The advanced capabilities of importing and exporting various 3D data made it possible to create virtual environments using the above-described sets of three-dimensional spatial data and their textures without great efforts. In addition, the ability to attach additional data to each of the objects in the form of separate data files allows you to implement an object-oriented approach and create a full-fledged model including the GIS functionality. This allows you to complement the graphic three-dimensional model with text, tabular, multimedia data, as well as hyperlinks to third-party pages, objects, applications.
In addition, UNITY provides the ability to start the system with predetermined parameters. For example, before starting the system, the position and point of the camera and its parameters can be specified, the environment parameters changed, etc. In this way, as part of testing, the test work of this model (Fig. 6) was carried out in the central part of Vladivostok in the mode of information support of the Russian rescue service 112 (analogue of USA 911).
Fig. 6. General view of the central part of Vladivostok. The complex spatial structure of the territory is clearly visible
5. PROJECTS, WHERE DESCRIBED APPROACH WAS USED
During 2017-2019 The described method was tested on a number of objects: the model of the village Domodedovo (Moscow region), model of the central part of Vladivostok (East Russia), model of the central part of Kaluga, model of the port territory Port Vera (East Russia). In all cases, self-running applications were created for use by a wide range of people, including non-specialists. The goals of creating each of the models are different. For example, the Domodedovo model was created as a technology demonstrator. The Port Vera model is intended for corporate use by employees of the company and for support of decision-making for the sustainable development of the port complex.
The model of the city of Kaluga is intended for use by the administration of the city of Kaluga for the purpose of providing information for urban planning and territorial planning. The model of Vladivostok was used to work out the interaction with the automated workplace.
As we see, the main (but not only) purpose of virtual models of territories is decision support and display of the urban environment for display to a wide range of people.
All models are created according to the principle of open architecture, and can be at any time supplemented by new attribute information, expanded both geographically and by the functional set or implemented into other systems as a component or called component.
CONCLUSION
New technological capabilities of collecting and geo-referencing spatial data, as well as visualization of three-dimensional information, make it possible to review the possibilities of creating and using virtual models based on an object-oriented approach. The developed methodological and technical solutions described in this article allow to create highly accurate virtual models of urban and industrial facilities that meet the requirements of modern realities in terms of functionality, detail and accuracy. The cost and time of creation, as well as the requirements for professional training of manufacturers of these models are significantly reduced compared to previously used.
Such an approach - the sharing of laser and photographic data - does not necessarily have to be tied to the use of a bundle of "terrestrial laser scanning + aerial photography with a UAV." Wide use of unmanned aerial systems equipped simultaneously with lidars and aerial cameras is on the way [2, p. 110]. They provide even more complete and detailed information about the objects (including information about the relief under the canopy of trees) - for example, the Riegl RiCopter is equipped with scanners with a capacity of up to 1.5 million points per second and cameras with a resolution of up to 50 megapixels.
In addition to quite obvious areas of use in the urban economy, a completely similar approach is possible in other areas of knowledge - when modeling archaeological objects, unique natural landscapes, areas of intensive environmental impact (open pits), as well as natural disaster zones (avalanche-prone areas, flood zones , areas of volcanism, etc.) [1, p. 146].
Further development of creation of virtual models will increase the functionality of spatial queries and updating the tools for modeling in 2.5D mode. The introduction of such models in the wide practice of geographical research is impossible without simplifying their production technology and increasing their
capabilities. It is possible to bring the functionality of the models to the level of basic GIS packages, which
finally can solve the problem of insufficient visibility of virtual models in GIS applications.
References / Список литературы
1. Allen Р.К., Stamos I., Troccoli A. A., Smith B., Leordeanu M., Hsu Y. 3D modeling of historic sites using range and image data // Proceedings of the 2003 IEEE International Conference on Robotics and Automation, 2003. 01, Pp. 145-150.
2. Chen Q. Airborne lidar data processing and information extraction // Photogrammetric Engineering & Remote Sensing, 2007. V. 73. № 2. Pp. 109-112.
3. Haala N., Brenner C., Anders K.-H. 3D urban GIS from laser altimeter and 2D map dat // International Archives of Photogrammetry and Remote Sensing, 1998. V. 32. Pp. 339-346.
4. Lohr U. Digital elevation models by laserscanning: Principle and applications // Third International Airborne Remote Sensing Conference and Exhibition, 1997. V. I. Pp. 174-180.
5. Schwalbe E., Maas H., SeidelF. 3D building model generation from airborne laser scanner data using 2D GIS data and orthogonal point cloud projections // Proceedings of the International Society for Photogrammetry and Remote Sensing, 2005. V. 3. Pp. 12-14.
6. Vosselman G., Dijkman S. 3D building model reconstruction from point clouds and ground plans // International Archives of Photogrammetry and Remote Sensing, 2001. V. 34. Pp. 37-43.
7. Zhang C., Chen T. Efficient feature extraction for 2D/3D objects in mesh representation // Proceedings of the 2001 International Conference on Image Processing, 2001. V. 3. Pp. 935-938.
