GeoChallenge Contest Winners - 2024

2024 ASPRS GeoChallenge Contest

3DEO® Inc. (3DEO) is proud to announce the results of its partnership with the American Society for Photogrammetry and Remote Sensing (ASPRS) in sponsoring the 2024 ASPRS Annual GeoChallenge. To increase awareness of the full potential of Geiger-mode (GM) lidar, 3DEO and ASPRS launched the competition designed to challenge university students and early career professionals to push the boundaries of innovation. The 2024 GeoChallenge evolved from the Geographic Information Systems (GIS) Day Map Contests of previous years, transitioning from a focus on map creation to exploring innovative applications of cutting-edge GM lidar data. For the competition, 3DEO provided three datasets of varying scene content. The scenes included the Charles Houston Shattuck Arboretum on the University of Idaho’s campus, an old New England mill town along the Merrimack River near Lawrence, MA, and Barrett Park in Leominster, MA.

First Place - Team Entry - Purdue University
Jinyuan Shao
Sungwoong Hyung
Stephanie Willsey
Sang-Yeop Shin
Aser M. Eissa
Hazem Hanafy

The 1st place submission was a team effort submitted by students from Purdue University. The team outlined a workflow to process and analyze GM lidar data, including denoising, semantic segmentation, and feature extraction. A tailored denoising method was developed to address the unique properties of GM lidar data. Semantic segmentation, using a deep learning model, classified features such as ground, vegetation, vehicles, and buildings. Ground and non-ground separation techniques produced digital terrain models (DTMs), while crown segmentation extracted vegetation attributes, including canopy height, area, biomass, and carbon content, across all three data sets. Results were made accessible via an interactive web portal for data visualization and interpretation.

2nd Place - Kris Brandt
Applied Geospatial Engineer

For his submission, Kris Brandt developed an automated pipeline utilizing the Point Data Abstraction Library (PDAL) to efficiently process and denoise GM lidar data, addressing the unique challenge of high noise levels inherent to the technology. Using PDAL, the script converts raw Binary Point Files (BPF) to LAS format, tiles large datasets for parallel processing, and applies statistical outlier filtering to remove noise while preserving critical key points of the model. By leveraging multiprocessing, the workflow ensures scalability for large-scale GM lidar collections, allowing for customization and fine-tuning.

3rd Place - David Abiola
Doctoral Candidate - Oregon State University

David Abiola developed an adaptive ground filtering tool to address challenges in lidar and point cloud data, such as distinguishing ground points from vegetation, managing varying terrain slopes, and reducing noise in large datasets. Accurate ground filtering is crucial for generating reliable DTMs used in applications like flood risk assessment and urban planning. The tool uses a grid-based analysis to segment lidar data, vertical difference filtering to refine ground points, and spatial filtering with Random Sample Consensus (RANSAC) to detect and adjust for ground points, adapting to different terrain types. Tested on GM lidar data from Leominster, MA, the tool demonstrated high accuracy and robustness, particularly in forested and urban environments, improving terrain representation and supporting reliable downstream analyses.