Aleksey Boyko will present his research seminar/general exam on Tuesday May 18 at 10 AM in Room 402. The members of his committee are: Tom Funkhouser (advisor), Szymon Rusinkiewicz, and Adam Finkelstein. Everyone is invited to attend his talk, and those faculty wishing to remain for the oral exam following are welcome to do so. His abstract and reading list follow below. ==================================== Abstract: Range scan data provides high resolution 3D information about the real life environment. Urban scans in the form of point clouds are becoming more available. However, point clouds are both noisy, very large, and they do not have semantic labels that associate each point of the scanned surfaces to a specific object. Our goal is to provide means for a meaningful classification of large scale point clouds derived from urban environment range scans. In my talk I will focus on two classes of objects. An important category of objects in an urban landscape consists of roads, streets and driveways. This category represents potentially the largest in size class of objects in a city - the road network. At the same time this class provides a context for other objects, both of a large and small scale - it separates the city into blocks and defines the position of large variety of small scale objects. In order to provide such context it is desirable to have a precise separation of the road surface from the sidewalks, which delimit the road surface. We present an algorithm that takes advantage of available online map information and uses dynamic contours (snakes) to provide an accurate extraction of points belonging to the road surface from the rest of the points. Addressing the rest of the objects we can separate them into large scale objects, such as buildings, and small-scale objects. Due to noisiness of the point cloud data it is harder to address the question of recognition of small-scale objects. [Golovinskiy, et al. 2009] offer numeric evaluation of how well supervised machine learning algorithms are at automatic classification of small objects in a point cloud of an urban environment. We research how making the classification user-guided can improve the rates and accuracy of the classification. We employ online learning (user-guided) and active learning (computer-guided) approach and show that giving user real-time control over the process of learning improves both the rate and the correctness of classification over blind selection of the training set. Reading list: [1] R. J. Campbell and P. J. Flynn. A survey of free-form object representationand recognition techniques. Computer Vision and Image Understanding, 81 doi: 10.1006/cviu.2000.0889 URL http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.89.1720&rep=rep1&type=pdf [2] M. Carlberg, J. Andrews, P. Gao, and A. Zakhor. Fast surface reconstruction and segmentation with ground-based and airborne lidar range data. 3DPVT, 2008. URL http://www.eecs.berkeley.edu/Pubs/TechRpts/2009/EECS-2009-5.pdf [3] A. Frome, D. Huber, R. Kolluri, T. Bulow, and J. Malik. Recognizing objects in range data using regional point descriptors. ECCV doi: 10.1007/b97871 URL http://www.springerlink.com/content/d4ukq18fpbfe9la6/fulltext.pdf [4] A. Golovinskiy, V. G. Kim, and T. Funkhouser. Shape-based recognition of 3d point clouds in urban environments. ICCV, Sept. 2009. URL http://www.cs.princeton.edu/gfx/pubs/Golovinskiy_2009_SRO/paper.pdf [5] D. Hearn and M. Baker. Computer Graphics with OpenGL. Pearson PrenticeHall, 3 edition, 2004. URL http://www.pearsonhighered.com/educator/product/Computer-Graphics-with-OpenG... 6.page. [6] A. Jaakkola, J. Hyyppa, H. Hyyppa, and A. Kukko. Retrieval algorithms for road surface modelling using laser-based mobile mapping. Sensors, 8, 2008. doi: 10.3390/s8095238 URL http://www.mdpi.org/sensors/papers/s8095238.pdf [7] A. E. Johnson and M. Hebert. Using spin images for effcient object recognition in cluttered 3d scenes. IEEE PAMI, 1999. URL http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.23.8816&rep=rep1&type=pdf [8] B. C. Matei, Y. Tan, H. S. Sawhney, and R. Kumar. Rapid and scalable 3d object recognition using lidar data. SPIE, 6234, 2006. doi: 10.1117/12.666235 URL http://spiedl.aip.org/getpdf/servlet/GetPDFServlet?filetype=pdf&id=PSISDG00623400000162340 1000001&idtype=cvips&prog=normal. [9] H. Mayer, I. Laptev, and A. Baumgartner. Multi-scale and snakes for automatic road extraction. ECCV, 1998. URL http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.74.8855&rep=rep1&type=pdf [10] R. Schnabel, R. Wahl, R. Wessel, and R. Klein. Shape recognition in 3d point clouds. The 16-th International Conference in Central Europe on Computer Graphics, Visualization and Computer Vision, 2008. URL http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.92.2344&rep=rep1&type=pdf [11] B. Settles. Active learning literature survey. 2009. URL http://pages.cs.wisc.edu/~bsettles/active-learning/ [12] S.P.Clode, F.Rottensteiner, P.Kootsookos, and E.Zelniker. Detection and vectorization of roads from lidar data. PhEngRS, May 2007. URL http://www.asprs.org/publications/pers/2007journal/may/2007_may_517-535.pdf [13] G. Vosselman and Z. Liang. Detection of curbstones in airborne laser scanning data. International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences, 38:3/W8, 2009. URL http://www.itc.nl/personal/vosselman/papers/vosselman2009.paris.pdf =========