Powered by Google’s Imagery Insights
Keen AI has developed Haystack™, a scaleable, automated system able to map utility infrastructure and connections using imagery, machine vision, and geospatial intelligence.
Haystack™ can rapidly generate accurate asset inventories without manual fieldwork. For this project, we used our unique access to Street View imagery through Google’s Imagery Insights to map overhead utility poles.
The Challenge Facing DNOs
Distribution Network Operators (DNOs) are under immense pressure to expand the grid to meet rising demand, support low-carbon technologies, and upgrade ageing infrastructure. However, they rely on incomplete and outdated asset records and often do not have full visibility of their network. Many poles, especially those carrying mixed assets like cables, transformers, lamps, and signage, are undocumented or inaccurately recorded.
Traditional field surveys are slow, expensive, and difficult to scale. To meet expansion targets and make better use of existing infrastructure, DNOs need a faster, better way to maintain a full inventory of their networks.
Deploying Haystack™: Automated Utility Pole Mapping
Haystack™ is a geospatial and machine vision system that uses imagery to automatically detect, classify, and map utility assets.
Haystack™ can address the urgent visibility challenge that DNOs are facing by creating a detailed, up-to-date asset inventory through a fully automated and scalable process. This reduces reliance on time-consuming and costly field surveys, enables faster and more consistent data collection across large areas, and simplifies future upgrades and maintenance.
- Automated Asset Detection
- High-Accuracy Geolocation
- Fast Deployment Within Weeks
- Cost-Effective & Scalable
- Image-Supported Complete Inventory
- GIS-Compatible Output
- Comprehensive Infrastructure Visibility
In a bespoke trial, Haystack™ was deployed to map overhead poles in a specified trial region, which delivered very strong results for the DNO:
Proven Results from a Real-World Deployment
1. Imagery Input
Haystack™ begins by intaking a combination of street-level and aerial imagery. While this trial used high-quality Google Street View images accessed through our unique partnership, the platform also supports customer-supplied or open-source images.
2. Detection, Geolocation & Classification
Using these images, Haystack™, powered by deep learning-based object detection and image segmentation models and combined with geospatial layers, was able to:
- Detect and geolocate poles with high accuracy
- Classify each pole as electric-only, lamp-only, or mixed-use
- Identify attached assets such as lamps, signs, and transformers
3. Connection Inference
Haystack™ then uses computer vision and spatial analysis to detect cable routes, estimate likely connections between poles, and infer connections from poles to nearby buildings.
The result is a single, GIS-ready dataset containing imagery, locations, classified assets, and inferred connections, designed for easy integration into existing systems for validation and asset management.
