Congratulations to Shiqian Wang

On Tuesday, April 26th, 2016, Shiqian Wang successfully defended her MES thesis entitled "Urban Roadside Tree Inventory Using a Mobile Laser Scanning System". The committee members are: Prof. Derek Robinson, from the Department of Geography and Environment Management, University of Waterloo;Prof. Michael Chapman, from the Department of Civil Engineering, Ryerson University; Prof. Yuhong He, from Department of Geography and Planning, University of Toronto, and the committee chair Prof. Jonathan Li, from the Department of geography and Environmental Management, University of Waterloo.

Thesis abstract:

Field surveys, as a traditional method for urban roadside tree inventory,is time consuming and labour-intensive, and, therefore, costly. However, more recent technology has allowed a vehicle-borne mobile laser scanning (MLS) system, driving at the speed limits on urban streets, to obtain three-dimensional (3D) georeferenced point clouds with high-density and high-accuracy. 3D MLS point clouds can be used to directly obtain3D information of roadside trees and extract geometrical information such as crown diameter, diameter at breast height (DBH : trunk diameter at 1.3 m height), and tree height. However, a variety of problems make the processing of MLS data very difficult: the large data volume, occlusions, mixed-density and irregular distribution of MLS point clouds as well as complexity of the road environment. Thus, automation of the MLS data processing is urgently needed.

The main goal of this thesis is to develop a method that can automatically extract urban roadside trees from the 3D point clouds acquired by a MLS system. The methods developed in this study consist of three parts. The first part deals with ground removal. As such, only off-ground points are used to extract trees. The second part handles tree detection by comparing four segmentation and clustering methods: the Euclidian distance clustering algorithm, the region growing segmentation method, the Normalized cut (Ncut) method, and thesupervoxel-based tree detection method. The third part focuses on automated extraction of tree geometric parameters such as tree height, DBH, and crown spread, and horizontal slices features.Finally, classification of tree species was conducted using the k-Nearest Neighbour (k-NN) and the random forests (RF) algorithm. A total of four MLS datasets (three in Xiamen, China and one in Kingston, Ontario) acquired in 2013 and 2015, respectively, were used to test the developed method. The ground truthing data of DBH estimation were obtained through manual measurement of selected roadside trees after the two MLS missions in Xiamen in the fall 2015. The field surveyed DBH values of the 163 roadside trees were used to estimate the accuracy of the proposed tree extraction method. The 200 manually labeled trees with 8 different species were selected to examine accuracy of the proposed classification method. The results show that over 90% of the roadside trees were correctly detected, with an average error of about 5% in DBH estimation when compared to the field survey, and an overall accuracy of 78% for the classification of tree species.