TH
T. Hemmes
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
Master thesis
(2018)
-
Tom Hemmes, Mathias Lemmens, P.J.M. van Oosterom, Kaixuan Zhou, Maarten Kruithof
There is a paradigm shift from two- to three-dimensional data, from maps to information dense models. Self-driving cars, digitization of historic buildings or maintenance of highway infrastructure are a small selection of many applications that use laser scanning to acquire three-dimensional data of our physical surroundings. Most of these applications require more than the shape of their surroundings. For example, a self-driving car needs to identify pedestrians, road signs and traffic lights in order to navigate safely. Therefore the point cloud acquired by laser scanning needs to be enriched with additional information. Automatic assignment of the object type a point belongs to is called classification.
This research focuses on deep learning for point cloud classification, because it revolutionized classification of imagery. PointNet is used to enable deep learning directly on point cloud data sets. To date PointNet is proven for indoor point clouds, this research explores the application of PointNet on an outdoor highway scene. The methodology creates point cloud training data efficiently by reusing known 2D object locations. Different spatial representations and sampling methods for the points are tested.
On average the method classifies 50% of the points correctly in four object classes. In combination with clustering of the point-wise predictions, the method predicts 60% of the 2D object locations successfully. The performance is comparable to the 47% average class accuracy PointNet achieves for 13 classes on the indoor data set. ...
This research focuses on deep learning for point cloud classification, because it revolutionized classification of imagery. PointNet is used to enable deep learning directly on point cloud data sets. To date PointNet is proven for indoor point clouds, this research explores the application of PointNet on an outdoor highway scene. The methodology creates point cloud training data efficiently by reusing known 2D object locations. Different spatial representations and sampling methods for the points are tested.
On average the method classifies 50% of the points correctly in four object classes. In combination with clustering of the point-wise predictions, the method predicts 60% of the 2D object locations successfully. The performance is comparable to the 47% average class accuracy PointNet achieves for 13 classes on the indoor data set. ...
There is a paradigm shift from two- to three-dimensional data, from maps to information dense models. Self-driving cars, digitization of historic buildings or maintenance of highway infrastructure are a small selection of many applications that use laser scanning to acquire three-dimensional data of our physical surroundings. Most of these applications require more than the shape of their surroundings. For example, a self-driving car needs to identify pedestrians, road signs and traffic lights in order to navigate safely. Therefore the point cloud acquired by laser scanning needs to be enriched with additional information. Automatic assignment of the object type a point belongs to is called classification.
This research focuses on deep learning for point cloud classification, because it revolutionized classification of imagery. PointNet is used to enable deep learning directly on point cloud data sets. To date PointNet is proven for indoor point clouds, this research explores the application of PointNet on an outdoor highway scene. The methodology creates point cloud training data efficiently by reusing known 2D object locations. Different spatial representations and sampling methods for the points are tested.
On average the method classifies 50% of the points correctly in four object classes. In combination with clustering of the point-wise predictions, the method predicts 60% of the 2D object locations successfully. The performance is comparable to the 47% average class accuracy PointNet achieves for 13 classes on the indoor data set.
This research focuses on deep learning for point cloud classification, because it revolutionized classification of imagery. PointNet is used to enable deep learning directly on point cloud data sets. To date PointNet is proven for indoor point clouds, this research explores the application of PointNet on an outdoor highway scene. The methodology creates point cloud training data efficiently by reusing known 2D object locations. Different spatial representations and sampling methods for the points are tested.
On average the method classifies 50% of the points correctly in four object classes. In combination with clustering of the point-wise predictions, the method predicts 60% of the 2D object locations successfully. The performance is comparable to the 47% average class accuracy PointNet achieves for 13 classes on the indoor data set.
Student report
(2017)
-
Tom Hemmes, Weiran Li, Jippe van der Maaden, Brenda Olsen, Marc-Julien Veenendaal, Stefan van der Spek, P.J.M. van Oosterom, Martijn Meijers, Theo Tijssen
Point clouds are becoming one of the most common ways to represent geographical data. The scale of acquisition of point clouds is growing steadily. However, point clouds are often very large in storage size and require computationally intensive operations. The integration of point clouds nowadays still face a lot of challenges. This project focuses on one of these challenges; integrating point clouds of different scales and granularity. Solving this challenge enables appealing visualisation, usability for low and high computation powers and geometrical consistency for analysis. The following question is researched: 'To what extent can a vario-scale approach improve integration of point clouds with varying point densities?'. A data model is created that uses importance as an additional dimension. This dimension contains an importance value which is calculated using two methods. Firstly random assignment of values to the points and secondly exact computed values. To compute this value the smallest distances to its nearest neighbour is assigned as importance value. A web application shows the results. Both random and exact methods show an exponential decay in distribution of the importance value. Though the random methods run much faster, the exact methods preserve much more edges and other details.
...
Point clouds are becoming one of the most common ways to represent geographical data. The scale of acquisition of point clouds is growing steadily. However, point clouds are often very large in storage size and require computationally intensive operations. The integration of point clouds nowadays still face a lot of challenges. This project focuses on one of these challenges; integrating point clouds of different scales and granularity. Solving this challenge enables appealing visualisation, usability for low and high computation powers and geometrical consistency for analysis. The following question is researched: 'To what extent can a vario-scale approach improve integration of point clouds with varying point densities?'. A data model is created that uses importance as an additional dimension. This dimension contains an importance value which is calculated using two methods. Firstly random assignment of values to the points and secondly exact computed values. To compute this value the smallest distances to its nearest neighbour is assigned as importance value. A web application shows the results. Both random and exact methods show an exponential decay in distribution of the importance value. Though the random methods run much faster, the exact methods preserve much more edges and other details.