AI/ML + Data Fusion + Infrastructure

The key word in describing our platform is AI-accelerated; a foundational principle of our service offering is that AI tools work best when they augment our staff’s extensive and technical expertise, not replace it.

What does this mean functionally? It means that in addition to leveraging our damage identification automation processes (which have been trained on over 10,000 miles of interstates and state routes), we have highway safety subject matter experts who physically put their eyes on the data we collect along highways; these experts ultimately make the call on damage identification and repair recommendations using industry standards and expertise. When false positives or false negatives are identified by our QA/QC staff, our on-going continuous improvement process ensures our tools are refined accordingly.

Additionally, our data collection operators follow established SOPs to ensure we get the best of both human and technological elements. This means that operators are trained to quickly and safely identify and log “at the push of a button” any asset damage they observe without stopping or exiting the vehicle. This provides an important supplemental quality check for our AI tools.

Great question! We recommend you check out our business case whitepaper to explore the hidden costs of hiring staff to perform technical condition assessment and defect identification work on the edge of dangerous, busy ROWs, especially for high AADT and geometrically constrained routes.

The short answer is yes, when you only account for the cost of an inspector's salary.

However, when you factor in other critical considerations (i.e., all-inclusive employment costs, equipment costs, IT deployment costs, liability and safety risks, the cost of staff training, comprehensive data management strategies, inconsistent inspection data quality, limited access to SME-level condition assessment resources, and more) the comparison becomes an apples-to-oranges exercise.

To summarize: manually inspections 1) come at a higher cost than is initially apparent, 2) don't guarantee route coverage, and 3) put DOTs at risk for the injuries/deaths that can occur when damaged highway safety assets are missed due to inconsistent inspection quality/turnaround times.

Our data collection processes leverage a combination of high frame rate, high resolution imagery cameras, GPS/IMU sensor fusion, and LiDAR scanners. Our multi-sensor hardware arrays are rated IP68 or above, which enables us to perform data collection and condition assessment irrespective of weather conditions.

Absolutely! A core part of our solution design philosophy is that infrastructure solutions are necessarily unique, based on the resources, needs, and management strategy of owner/operator.

We aren't just a data or technology company. Our staff and management team consists of experienced engineers, project/program managers, and infrastructure repair/replacement experts who've dedicated their careers to creating and improving the built environment.

You are probably right! That’s where our solution differs: Guardian Angel’s AI tools don’t rely on a single type of computer vision or analytical model. Our technology stack is multi-modal and multi-dimensional.

Some examples:·       
- Every point or image collected is geo-referenced to centimeter accuracy; that means every point or image (and corresponding asset defect) our platform detects can be precisely located by maintenance contractors or staff. ·       
- Parametric & geometric analysis is leveraged for semi-automated feature extraction from high density LiDAR points clouds.·        
- PointCNN machine learning models - trained on thousands of miles of real, regional highway data – are leveraged for augmented point cloud classification and highway feature extraction (road surface, road markings, highway safety assets like guardrail and cabling, signage, vegetation encroachment, and more)·        
- State-of-the-art computer vision models are leveraged for frame-by-frame defect detection analysis of imagery datasets.