Deep Learning with LiDAR Point Cloud Data for Automatic Roadway Health Monitoring

Term Start:

November 1, 2023

Term End:

May 31, 2024




Cloud Data, Deep Learning, LiDar, Roadway Health Monitoring

Thrust Area(s):

Data Modeling and Analytic Tools

University Lead:

California State Polytechnic University, Pomona


Yongping Zhang

Traditional methods for monitoring road conditions are fraught with challenges. Field inspections are labor-intensive and costly, aerial photography is subjective, and mobile measurement systems (MMS) require substantial investment in geospatial technology. In response to these limitations, there is a growing interest in leveraging advanced 3D scanning technologies, such as LiDAR and RGB-D scanners, in conjunction with deep learning algorithms for infrastructure assessment. 3D point cloud data, analyzed through deep learning models, offers several advantages over traditional 2D-based computer vision techniques. These include enhanced spatial resolution, superior object recognition, and the ability to handle complex scenes more effectively. However, this approach also introduces challenges, such as greater computational demands and the need for specialized hardware. Therefore, albeit with the tremendous benefits associated with the 3D point cloud, there are very few studies dedicated to the application of 3D point cloud-based deep learning models to the infrastructure operation and assessment. To bridge this research gap, this study aims to investigate the efficacy of various point cloud-based deep learning models in automating roadway health assessments.

Given the vital nature of this topic, the research will evaluate promising deep learning architectures, such as PointNet, PointNet++, 3D-CNNs, and PointCNN, using point cloud data gathered from multiple roadways in Southern California. Additionally, some typical technical challenges such as the noise filtering, data alignment, and dimension reduction via resampling, etc., will be further explored. This investigation aims to offer valuable insights into the pros and cons of these models under diverse conditions, thereby contributing to future research in this emerging area. Most importantly, these applications combine to offer a more comprehensive, real-time understanding of roadway health, facilitating proactive maintenance, reducing costs, and improving public safety.

Scroll to Top