Precision Annotation on Road and Infrastructure Damage

Disclaimer: Aya Data holds to the confidentiality agreements made with our clients. We shall not disclose any confidential information regarding clients and/or projects, pursuant to the individual agreements made. Consequently, client and project names and any identifiable information may be kept anonymous in our Case Studies.

Overview

The Client is a leading utility company that works to increase the longevity and improve the safety of infrastructure by utilizing cutting-edge tech solutions. The Client engaged its in-house data science team to develop a computer vision model that should detect and classify road cracks, which would enable efficiency and timely maintenance.

The solution would use drone and satellite imagery as well as photographs captured by ground-based vehicles. Ultimately, the CV model should provide engineers and maintenance crews with the information needed to prioritize repair efforts and enhance road safety.

Data Annotation

Image segmentation

Industry

Utilities

Headquarters

London, UK

Company Size

250+

Challenge

The Client contacted Aya Data to annotate 250,000 images of various road surfaces and crack types with bounding boxes, polygons, and semantic masks. The annotated data would be used to train the Client’s computer vision model.

As the project progressed, it became apparent that:

  1. because of the varying nature of road cracks and their potential impact on infrastructure, the Client would need a higher degree of annotation accuracy than was initially projected
  2. the solution needed to be scalable as the volume of images was increasing and the captured images were under different lighting and weather conditions, which needed to be considered during the processing.

Contact us

Solutions

We addressed these challenges by:

Enhancing annotation accuracy: We implemented a rigorous human-in-the-loop quality control mechanism to ensure that all annotations meet the required level of precision. Additionally, all of our annotators received specialized training that focused on accurately identifying and classifying various road crack types.

Scaling the solution: Aya Data optimized its annotation platform so that it could handle the scaling volume of images without losing the required level of annotation accuracy. We also adapted the platform so that it could process photographs under different lighting and weather conditions, in order for the annotation to remain consistent and reliable.

Results

In part because of our involvement, the Client successfully developed an accurate computer vision model that could detect and classify road cracks, enhancing the efficiency of their maintenance crews by enabling timely and targeted repairs. Ultimatelly, this improved road safety and reduced the potential impact road cracks could have on infrastructure.

Additionally, the time and resources that were saved due to Aya Data’s scalable annotation solution allowed the Client to further refine their CV model.