SOURCE: Tetra TechDESCRIPTION:
Dr. Reza Malehmir used artificial intelligence (AI) to fully automate data processing for a large provincial highway asset classification project, resulting in significant cost and time savings.
The Tetra Tech team planned to collect LiDAR data for a more-than-7,500-mile roadway network, then use commercial software and teams of highly trained personnel to extract details about and classify assets such as traffic signs. When using that approach could not meet the client’s required timeline within budget, the team looked for an alternative that would meet those needs and Tetra Tech’s rigorous quality requirements. Reza proposed automating the process using Tetra.Analytics, one of our foundational Tetra Tech Delta technologies.
Using AI to fully automate asset data extraction
Reza used machine learning algorithms to automate asset extraction and classification for our solution. He trained the AI models—using previously collected roadway data for the algorithm to learn from and maximize prediction success—and completed the first functional model in less than two months. Employing an iterative process, he spent evenings and weekends over several months developing the algorithms to fully automate asset data extraction and classification. The system can collect data on more than 60 unique assets, provides more accurate data than manual extraction, and can even identify assets almost fully obscured from human view.
The team was able to switch from manual to automated data extraction on the third of eight project deliverables, providing the final deliverables ahead of schedule. Thanks to continued improvements in the model’s accuracy and efficiency, the deliverables exceeded the client’s expectations.
“Not only did the move to an automated solution reduce total person-hours, it also produced a significantly faster delivery,” according to David Firbank, manager of Tetra Tech’s Roadway Infrastructure Data Collection Technologies group. “Over time, the automated extraction has become faster and more accurate than the best human extractors.”
Using this solution, Tetra Tech can deliver automated extraction of assets more quickly and at significantly lower cost than competitors who use manual extraction. Several of the assets extracted have become standard deliverables for some of our largest clients. The technology not only reduced production time on this project but also will reduce production time by thousands of person-hours on future projects.
Extending the impact of the technology
"With Tetra Tech’s domain expertise in various engineering fields, we are capable of creating machine learning solutions rapidly and efficiently that differentiate us from our competitors," said Reza.
Reza expanded the technology beyond its initial use by adapting the algorithms to support real-time object detection and classification from LiDAR data processing for Tetra Tech’s RailAI™ rail track assessment platform. He also modified the models to identify assets in digital images and video from GoPro cameras. This capability provided a flexible, low-cost data collection solution that can replace traditional, time-consuming manual assessments of linear asset classes such as gravel roads, sidewalks, and trails.
We also developed our own crack detection algorithm to augment third-party automated road condition software, improving accuracy and reliability and decreasing cost compared to competitors using the industry-standard data acquisition system.
Additional proposed uses, including automated vertical building façade assessment applications, represent innovations not currently available in the industry.
Congratulations to Reza for transforming a data collection challenge into a competitive, flexible, and low-cost solution for Tetra Tech’s clients.
KEYWORDS: Tetra Tech