ZC
Z. Chen
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
2 records found
1
Three-dimensional building models play a pivotal role in shaping the digital twin of our world. With the advance of sensing technologies, unprecedented data acquisition capabilities on capturing the built environment have surfaced, with photogrammetry and light detection and ranging being the two important sources, both of which can acquire point clouds of buildings. A point cloud is anisotropically distributed in space, which---though conveying spatial information itself---has to be converted into a surface model for a wider spectrum of usage. This conversion is often referred to as reconstruction. Despite the enhanced availability of point cloud data in the built environment, how to reconstruct high-quality building surface models remains non-trivial in remote sensing, computer vision, and computer graphics. Most reconstruction methods are dedicated to smooth surfaces represented by dense triangles, irrespective of the piecewise planarity that dominates the geometry of real-world buildings. Although some works claim the possibility of reconstructing piecewise-planar shapes from point clouds, they either struggle to comply with specific geometric constraints, or suffer from serious scalability issues. There is no versatile solution yet for building reconstruction. In this thesis, we propose a novel framework for reconstructing compact, watertight, polygonal building models from point clouds. Our approach comprises three functional blocks: (a) a cell complex is generated via adaptive space partitioning that provides a polyhedral embedding as the candidate set; (b) an implicit field is learnt by a deep neural network that facilitates building occupancy estimation; (c) a Markov random field is formulated for surface extraction via combinatorial optimisation, where an efficient graph-cut solver is applied. We extensively evaluate the proposed method in comparison with state-of-the-art methods in shape reconstruction, surface approximation and geometry simplification. Experimental results reveal that, with our neural-guided strategy, high-quality building models can be obtained with significant advantages over fidelity, compactness and computational efficiency. The method shows robustness to noise and insufficient measurements due to occlusions, and generalise reasonably well from synthetic scans to real-world measurements. Moreover, our method remains generic to not only buildings, but any piecewise-planar objects.
...
Three-dimensional building models play a pivotal role in shaping the digital twin of our world. With the advance of sensing technologies, unprecedented data acquisition capabilities on capturing the built environment have surfaced, with photogrammetry and light detection and ranging being the two important sources, both of which can acquire point clouds of buildings. A point cloud is anisotropically distributed in space, which---though conveying spatial information itself---has to be converted into a surface model for a wider spectrum of usage. This conversion is often referred to as reconstruction. Despite the enhanced availability of point cloud data in the built environment, how to reconstruct high-quality building surface models remains non-trivial in remote sensing, computer vision, and computer graphics. Most reconstruction methods are dedicated to smooth surfaces represented by dense triangles, irrespective of the piecewise planarity that dominates the geometry of real-world buildings. Although some works claim the possibility of reconstructing piecewise-planar shapes from point clouds, they either struggle to comply with specific geometric constraints, or suffer from serious scalability issues. There is no versatile solution yet for building reconstruction. In this thesis, we propose a novel framework for reconstructing compact, watertight, polygonal building models from point clouds. Our approach comprises three functional blocks: (a) a cell complex is generated via adaptive space partitioning that provides a polyhedral embedding as the candidate set; (b) an implicit field is learnt by a deep neural network that facilitates building occupancy estimation; (c) a Markov random field is formulated for surface extraction via combinatorial optimisation, where an efficient graph-cut solver is applied. We extensively evaluate the proposed method in comparison with state-of-the-art methods in shape reconstruction, surface approximation and geometry simplification. Experimental results reveal that, with our neural-guided strategy, high-quality building models can be obtained with significant advantages over fidelity, compactness and computational efficiency. The method shows robustness to noise and insufficient measurements due to occlusions, and generalise reasonably well from synthetic scans to real-world measurements. Moreover, our method remains generic to not only buildings, but any piecewise-planar objects.
SCIPoC: Semantic Classification of Indoor Point Cloud
A study into the possibilities of classifying indoor point cloud using a Deep Learning approach
Student report
(2020)
-
M. Smit, Z. Chen, M.A. Erbaşu, Y.A.L. Gaol, X. Li, E. Verbree, B.M. Meijers, J. Balado Frías, N. van der Vaart, R. Bunder
With the constantly evolving range of applications for technology the quality and amount of data constantly increases as well. In this growing data environment, there is a constant search to provide more value to all data that is available for as little effort as possible. Our research tries to add such additional value by diving into the concept of classifying point cloud by using deep learning, specifically in the indoor environment. This is done by first doing a neural network comparison and then doing a case study. In the neural network comparison, a look is taken into which of the neural networks that are capable of working with point clouds is best suited for our experiments in the indoor scene, based on the training speed, accuracy, ease of use concerning training on external datasets and setting up the network and space efficiency. After the comparison, we chose to continue with the PointCNN network during the case study. The case study is performed on data the NS (Nederlandse Spoorwegen) provided to us and all test results we got from our experiments can be visualized using the web application we developed along with this project. The purpose of the case study is to add extra value to the indoor LiDAR point cloud the NS has captured from Amersfoort Station by using deep learning to automatically classify assets present in their data. The value is in purposes, such as asset management, where the data does not need possibly hundreds of man-hours to be labelled. This saves a lot of time and also money each time a scan is made. In the case study we found through 4 different experiments that unbalanced data makes for bad results, but when a scene is labelled correctly very good results can be found in a local scene.
...
With the constantly evolving range of applications for technology the quality and amount of data constantly increases as well. In this growing data environment, there is a constant search to provide more value to all data that is available for as little effort as possible. Our research tries to add such additional value by diving into the concept of classifying point cloud by using deep learning, specifically in the indoor environment. This is done by first doing a neural network comparison and then doing a case study. In the neural network comparison, a look is taken into which of the neural networks that are capable of working with point clouds is best suited for our experiments in the indoor scene, based on the training speed, accuracy, ease of use concerning training on external datasets and setting up the network and space efficiency. After the comparison, we chose to continue with the PointCNN network during the case study. The case study is performed on data the NS (Nederlandse Spoorwegen) provided to us and all test results we got from our experiments can be visualized using the web application we developed along with this project. The purpose of the case study is to add extra value to the indoor LiDAR point cloud the NS has captured from Amersfoort Station by using deep learning to automatically classify assets present in their data. The value is in purposes, such as asset management, where the data does not need possibly hundreds of man-hours to be labelled. This saves a lot of time and also money each time a scan is made. In the case study we found through 4 different experiments that unbalanced data makes for bad results, but when a scene is labelled correctly very good results can be found in a local scene.