Discover, Connect & Collaborate at TECHINNOVATION 2021

Infrastructure Anomaly Detection Utilizing Machine Learning

Technology Overview

This solution is one among numerous AI and Machine Learning solutions developed at our Data & AI Center of Excellence. The solution automates manual inspection and monitoring of pavement and other infrastructure by identifying anomalies such as cracks and other damages. The solution is powered by Machine Learning, which not only identifies anomalies accurately, but also excludes occlusions, shadows, and other artifacts. It is a proprietary developed Computer Vision and Machine Learning algorithm to identify anomalies, and it has greatly reduced the resources necessary to conduct inspection and monitoring in areas such as pavement inspection, where the effort to inspect per ~1 hectare was reduced from 1 week to 1 hour.

Inspection of infrastructure, such as pavement, is a time and human resource heavy task. By providing the inspectors with automated reports and insights on the areas to be surveyed, organizations can optimize their human resources to focus on value adding tasks and complex decision making.

The target user of our solution includes Smart City solutions providers, infrastructure maintenance service providers, construction, property and real estate developers, drone manufacturers and/or service providers. We are looking for industry partners and/or technology partners to collaborate to prove the feasibility of the solution in the region.

Technology Features, Specifications and Advantages

  • Automating infrastructure inspection is a difficult task due to occlusions, shadows and noise in the images (e.g. road signs). By applying our proprietary Machine Learning models on this problem, we have managed to circumvent these previous limitations. The technology enables organizations to utilize a data driven approach and reduce its dependency on the tacit knowledge of their inspectors and subject matter experts. Additionally, our solution allows the organization to gather valuable data points over time, providing further prediction and insights that was previously impossible.
  • Over the last decade Machine Learning has proved to be dominant over traditional deterministic solutions. Most of pavement inspection methods only utilizes a few random samples from all existing defects due to time and man-power constraints, and manual measurement is often estimated by some simplification and estimation techniques which is usually inaccurate. Our solution improves the efficiency and productivity of inspectors by automating a traditionally human resource heavy process. It provides them with the right information at the right time to support them to make more accurate decisions, which can significantly decrease the time of a survey, as the inspectors can focus on the areas that are of importance instead of having to survey the area in person or view all footage personally.
  • Applying Computer Vision and Machine Learning models to optimize infrastructure maintenance is a novel approach that we have successfully proven to greatly increase productivity. It is easily scalable due to the nature of computer vision and machine learning technologies, while also removing the need for the inspector to be on the site, allowing them to only focus on high value adding tasks and complex decision making. Our solution complies with the ASTM standard for pavement evaluation has been fulfilled and was even further improved.

Potential Applications

  • The primary application is to automatically identify anomalies, such as cracks and other damages, in infrastructure and provide inspectors with the necessary insights to support their decision making at a fraction of the time. It can automate any inspection process that is visible with top-down drone imagery and on non-reflective surfaces.
  • Potential product areas include Machine Learning powered infrastructure inspection platform utilizing computer vision to identify anomalies and damages and track them over time. The data gathered can be utilized to develop future predictive maintenance models, which can provide organizations with data driven predictions and insights to drastically improve their operations.
  • With the current Inspection, Repair and Maintenance market (IRM) in APAC having reached US$9.27 billion in 2019, there is a significant opportunity for our solution in the region.

Customer Benefit

  • Significantly lower the time for large scale inspections (reduce pavement inspection effort from 1 week to 1 hour per ~1 hectare).
  • Significantly reduce human effort for inspection, allowing inspectors to focus only on high value adding tasks and complex decision making.
  • Provide a data driven approach to decision making, reducing the tacit knowledge dependency of inspectors.
Contact Person

David Bergendahl


Crayon Pte. Ltd.

Technology Category

  • Green Building
  • Building Automation / Management
  • Infocomm
  • Artificial Intelligence, Smart Cities, Video/Image Analysis

Technology Readiness Level


Computer Vision, AI, Machine Learning, Inspection, Maintenance, Anomaly Detection