AI-Based Feature Extraction in TBC Adds Value to Collected Data
The artificial intelligence (AI) built into Trimble® Business Center software (TBC) expedites tedious, repetitive activities such as point cloud classification and feature extraction.
By matching software that enables automated feature extraction with hardware that rapidly collects high-quality data, Trimble creates a complete solution in one efficient package. Whether managing assets, inspecting roads, creating digital terrain models (DTM), or monitoring stockpile volumes at construction or mining sites, AI helps deliver accurate information faster than ever before.
Automated classification and feature extraction workflows in TBC are based on a combination of AI techniques, including 3D and 2D deep learning. Using multiple AI methods provides flexibility and allows Trimble to deliver outputs with premium quality.
Classification capabilities are based on a 3D deep learning semantic segmentation model that is trained on representative datasets showing a broad range of geographic locations to ensure coverage of as many user groups as possible. The software performance continues to improve over time as it learns to recognize more key characteristics of the many features.
A mix of AI techniques is packaged in a user-friendly streamlined interface for automated feature extraction. Feature extraction commands generate the vectors and geometric attributes for each asset in user point clouds, such as powerlines, manhole covers, trees, pavement markings, poles and signs. With new stockpile extraction and pavement inspection functionalities, Trimble introduces complex AI-based workflows that automate data extraction and data analysis.
For added flexibility, a new tool for training 3D deep learning models allows users to customize feature extraction workflows to serve individual domain-specific needs.
Customized Feature Extraction
GeoVerra, Canada’s most established land surveying and geomatics firm, works with clients to support their projects with their industry experts and leading-edge technology in a range of industries including transportation, municipalities, telecommunications, and oil and gas.
TBC with AI capabilities is the standard software used to provide two primary services:
- Performing feature extraction from point clouds/imagery collected with mobile mapping
systems, scanners and drones. - Creating deliverables such as CAD and 3D files, pavement assessments, databases and GIS. TBC with AI provides a streamlined workflow with excellent results.
Mobile mapping data is classified with 3D deep learning in TBC. Photo Credit: GeoVerra.
“We see at least a 30 per cent time savings on large projects when performing feature extraction and classification in TBC,” says GeoVerra’s National Manager, Mobile Solutions, Alex Garcia. “The comprehensive information generated adds value and helps us meet and sometimes exceed customer expectations.”
Garcia adds, “Once we process mobile mapping data we classify immediately and send the appropriate files to people just starting the project. We only send the necessary parts to the correct people, for example, powerlines to the people who are working with powerlines, so everyone is more efficient.”
Custom 3D deep learning models are trained to extract features such as curbs, trash cans, fire hydrants, traffic lights, lamps on poles and sign plates using TBC and are applied on the top of generic classification. Photo Credit: GeoVerra.
GeoVerra uses TBC as their standard software for point cloud classifications for municipal asset management and for automated extraction of many typical features, such as powerlines, manholes, poles, and curbs.
Curb classification and extraction in TBC. Photo Credit: GeoVerra.
Using the new training tool, Garcia trained GeoVerra’s 3D deep learning models to automatically classify curbs, sidewalks, trash cans, etc., throughout municipalities, saving considerable time compared to a manual approach. Training of custom deep learning models is available in: Extract Classified Point Cloud Regions command > Training tab in TBC.
GeoVerra uses Trimble’s complete workflow for automated pavement condition inspection to identify potholes, bumps, depressions, corrugation, rutting and cracks in the mobile mapping data. This calculates the pavement condition score according to internationally recognized American Society for Testing and Materials (ASTM) standards.
Pavement Inspection functionality in TBC automatically extracts distresses and calculates Pavement Condition Index. Photo Credit: GeoVerra.
Thanks to improved productivity, GeoVerra can take on more business during the busy season in Canada (mid-spring to mid-fall) before winter weather interferes with mobile mapping. The team handles more projects in a shorter time with improved efficiency and reduced processing time.
Garcia says, “I like where TBC is going—the capability to process mobile mapping and UAV data and bring in scan data from all types of hardware eliminates switching back and forth between different software programs. In addition, feature extraction using AI in TBC is a great time saver. Customers are surprised how fast we can turn around data.”
Earthworks Site Management
Severino Trucking, an earthworks site contractor based in New Hampshire, uses TBC for site modeling and quantifying stockpiles on construction sites for improved safety and shorter turnaround of deliverables. With project managers expecting information faster, the AI capabilities in TBC allow Severino Trucking to complete same-day processing of drone imagery for stockpile extraction and surface measurements, delivering topographic maps within a day instead of weeks.
Automatic point cloud classification removes trees, people and equipment from UAV point clouds to provide a bare ground dataset. Photo credit: Severino Trucking.
For site modeling, a typical site is between 30–60 acres in size. Automatic point cloud classification based on 3D deep learning in TBC involves removing noise such as trees, people and equipment. With AI doing the labor-intensive classification work, Severino estimates an 85 per cent reduction in the time it takes to create a DTM.
“The use of AI allows us to adjust our pace to meet the needs of our customers,” says Severino Trucking’s Project Engineer, Pat L'Heureux. “With a UAV, we can do multiple flights in the morning, process the data with TBC in a couple hours and create deliverables—either design or quantification—the same day. AI turns a multi-day process into a single-day process which helps keep projects moving forward.”
Technology simplifies stockpile management and reduces the need for repetitive manual tasks. Photo credit: Severino Trucking.
Severino Trucking collects aerial imagery with a drone for stockpile quantification, usually conducted monthly. In 20 minutes, the drone captures better data than traditional methods, including a background image of the surrounding area, and the process is safer than a boots-on-the-ground survey. The stockpile extraction tool simplifies the process of managing stockpiles. L’Heureux says, “I used to have to create multiple surfaces and cut/fill maps, but the new user interface makes it quite easy.”
AI technology is not replacing Severino Trucking’s people, it’s improving the quality of life by reducing repetitive tasks. Rather than eliminating the jobs of the layout workers, AI removes the monotonous part while delivering information faster.
"Workers appreciate the improvements and benefits,” says L’Heureux. "We can focus on design work and spend more time analyzing data, which leads to smarter and quicker operational decision making."
Stockpile Measurements from Point Clouds
LE34, Denmark’s largest provider of surveying and land management services, uses a comprehensive suite of Trimble solutions, including TBC software, at 29 offices throughout the Nordic region.
With the release of TBC 5.90, the simplified stockpile extraction feature automatically extracts stockpile boundaries, calculates volumes, and creates reports with volume, date, slope area and base area with limited manual interaction necessary. To calculate stockpile volumes in TBC go to Point Clouds tab > Extract Stockpile. The command is executed by selecting the general location of the stockpile in the point cloud and TBC takes care of the rest.
Automated stockpile extraction and calculations in TBC streamline the workflow.
Photo credit: LE34.
René Bundgaard Christensen, Land Inspector at LE34, describes one project: “We collect point clouds of around 350 stockpiles of potting soil three times each year for a customer who used to manually measure their materials. Now using drones and TBC, we calculate the ground control points and classify the point clouds much more quickly. The automated stockpile extraction and calculations in TBC take about half the time compared to manual calculations.”
TBC greatly impacted the typical workflow because in one click the boundary is drawn at around 80–90 per cent accuracy, requiring only a small amount of editing. The potting soil stockpiles are flown in one and a half days, with data processed in a day and stockpile calculations completed in less than a day. This information can be exported as either a TBC report or directly to a CSV file.
Automated feature extraction and classification in TBC enables workers to identify, manage, and analyze any stockpile in a point cloud quickly and accurately.
Christensen also used TBC’s automated classification feature at a harbor to document conditions before and after a construction project. From 2,500 drone pictures, he created a point cloud and classified everything from buildings to poles to vegetation. For each visit, data collection and processing took one day, and just one hour to classify.
“Artificial intelligence is having a huge impact on the work we do, specifically by reducing the time it takes to extract features and classify point clouds in TBC,” says Christensen. “The modules are easy to learn. I look forward to using the TBC training module to extract custom features in the future.”