3D TREES: Tree Recognition via fEature Extraction & Segmentation

• Lindsey Wang, Electrical Engineering, University of Delaware

• Kenneth Barner, Electrical & Computer Engineering, University of Delaware

As a rising sophomore, my summer research is an extension of my involvement in the ECE department’s VIP (Vertically Integrated Projects) Drones and 2023 Winter Fellows program. The goal is to develop a three-dimensional (3D) UD campus model that allows for feature extraction, segmentation, and 3D printing. UD’s Mitchell Hall and general surrounding area have been selected for this research. On the hardware side, the DJI Matrice 350 RTK drone and Zenmuse L1 LiDAR scanning system have been employed for data collection; on the computing side, an Ubuntu virtual machine with various C++ based tools/libraries such as Point Cloud Library (PCL) has been set up for the 3D modeling effort. Overall, an enhanced 3D UD campus can be viewed from any angle, which can greatly supplement the visiting/viewing experience. Additionally, it serves as a baseline for change detection analysis and identification of potential campus enhancement.

Five sequential objectives were set for my summer research: 1) review existing literature related to feature extraction and segmentation; 2) understand the general format of point cloud data (PCD) and PCL; 3) select tools that facilitate the transformation and visualisation of LiDAR data, especially LAS (laser) and PCD; 4) set up a Linux environment to experiment on LAS/PCD data; and 5) enhance the 3D LiDAR model (Mitchell Hall) through tree extraction and segmentation for better visualisation and 3D printing based on previous steps.

I collaborated with my graduate student mentor to design a flight mission and collect LiDAR data for Mitchell Hall. I have successfully completed the following: 1) data collection; 2) preprocessing through tiling and downsampling for efficient processing and visualisation; and 3) identifying tree branches in the Mitchell Hall terrain point cloud data by modifying and implementing Andrew Burt’s tree segmentation algorithm.