استخراج و مدلسازي ساختمان با استفاده از تصويرسازي ابر نقاط ليدار در صفحات دوبعدي

Building Extraction and Modeling Using LiDAR Point Clouds Imaging on Two-Dimensional Surface

امروزه ارائه مدل سه بعدي از عارضه هاي جهان واقعي بسيار مهم و پركاربرد است و توجه محققان در شاخه هاي مختلف ازجمله نقشه برداري و سيستم هاي اطلاعات مكاني و علاقه‌مندان به بازسازي سه بعدي ساختمان را به خود جلب كرده است. ساختمان كليدي ترين بخش اطلاعات در يك مدل سه بعدي شهر است، ازاين‌رو استخراج و مدلسازي ساختمان ها از داده هاي سنجش از دور گام مهمي براي ساخت مدل رقومي يك شهر محسوب مي شود. در اين مقاله، روشي جديد براي بازسازي ساختمان هاي مناطق شهري در مجاورت با پوشش گياهي از ابر نقاط نامنظم ليدار ارائه مي شود. اين روش با بهره گيري از استراتژي داده مبنا شامل دو مرحله؛ استخراج نقاط ساختماني و مدلسازي ساختمان است. در مرحله اول نقاط پوشش گياهي اوليه با استفاده از پالس هاي بازگشتي ليدار و نقاط زميني با استفاده از روشي بخش مبنا حذف مي شوند. سپس نقاط سقف ساختمان با استفاده از تصوير سازي نقاط در صفحات دو بعدي XY، XZ و YZ با كنار گذاشتن نقاط پوشش گياهي متراكم استخراج مي شوند. در مرحله دوم لايه هاي ارتفاعي سقف ساختمان جداسازي شده و رئوس مهم سقف هر لايه با برازش خط و سرشكني راستاي خطوط استخراج مي شوند. درنهايت پس از تشكيل مدل سقف ساختمان بااتصال رئوس مهم سقف، با اضافه كردن ديوار هاي ساختمان مدل كامل ساختمان ايجاد مي گردد. در اين تحقيق تمركز بر روي مدلسازي ساختمان هاي در مجاورت پوشش گياهي است. اين روش بر روي ساختمان هايي با سبك هاي مختلف معماري سقف و اندازه متفاوت پياده سازي شد و بيشتر ساختمان ها با موفقيت مدلسازي شدند؛ اگرچه با افزايش پوشش گياهي و جزئيات ساختمان از دقت مدلسازي كاسته مي شود. ساختمان هاي بازسازي شده به‌طور متوسط با دقت 42 سانتي متر براي نقاط گوشه ساختمان مدلسازي شده اند.

Nowadays the three-dimensional presentation of real world features is very important and useful, and attracted researchers in various branches such as photogrammetry and geographic information systems and those interested in three-dimensional reconstruction of the building. Buildings are the most important part of a three dimensional model of a city, therefore extraction and modeling buildings of remote sensing data are important steps to build a urban digital model. Although many efforts to reconstruct the three-dimensional building of LiDAR data have been made by researchers in recent years, but challenges still exist in this area, especially in urban areas. In previous studies, dense vegetation and tall trees in the vicinity of the buildings cause to difficulty in the building extraction process and reduction in the accuracy of the modeling results. The aim of this article is extraction and reconstruction of buildings by using LiDAR point clouds in urban areas with high vegetation. In this study, factors such as the LiDAR return pulse, the height of points and area of the region is used to separate the non-structural parts. Ground points in segment-based method by changing the size segment in each iteration by mean and standard deviation of the height of points in any segment extracted. The vegetation points are extracted and identified using LIDAR return pulse and a new method called three-dimensional imaging of points on two-dimensional surfaces . The projection process is done in the planes of XY, XZ and YZ. Using Illustration of points and changing the angle of view makes the point clouds be evaluated in different directions. Region expanding algorithm and length constraints imposed in different planes has an important role in the separation of dense vegetation. The modeling of building is done by using break lines and important vertices of the building roof in layers of roof height. Extraction of building edge points and height layers of roof is done separately in each building. This points are isolated by height analysis of the roof points. In the line approximations grouping the points in each height layer, line fitting and adjustment of line directions are factors that caused the break lines and points of the building roof to be correctly created. In roof modeling, the basic structure of the roof is modeled and then the parts on the roof are added to the model. The overall structure of the roof is made by roof vertexes and normal vector of generated planes. At the end, by calculating the point’s distances from roof plane, the roof parts are identified and the model of this components are added to the roof. The proposed method is evaluated on LiDAR point clouds in an area of the Stuttgart, Germany, with a density of 4 points per square meter. The accuracy of the proposed method is evaluated by visual interpretation and quantitative comparisons with information extracted by a human operator. The accuracy of proposed method is about 96 percent in extracting building points and modeling error at the corner of the building is approximately 44 cm. Overall, the results represent the success of the proposed method in extracting and modeling of buildings in areas with dense vegetation.