Annotation of pulling chambers
Find out how we helped Alteia by annotating pulling chambers to train their model that detects urban telecoms networks from aerial images.
Must-read Case Studies
Foodtech App
3D Point Cloud Annotation Tool
Aerial images
Medical 3D annotation
Microscopic annotation
Sports analytics
...and many others case studies that are confidential!
All too often, roadworks cause significant damage to sewerage, water, electricity, telecoms and gas networks, resulting in substantial additional costs and sometimes serious accidents. Poor knowledge of the location of these networks is the main cause of such damage.
Alteia offers network operators and local authorities a robust and effective solution for detecting and accurately locating network infrastructures.
Context
CONTEXT
Alteia uses AI to map its customers' underground networks.
Pulling chambers are underground cavities accessible through a trapdoor, designed to facilitate the pulling of cables through conduits buried in the ground.
To make it easier to localize infrastructure elements, Alteia uses AI and Computer vision model. Once localized, these infrastructure elements are fed into a French standardized geographical repository known as the PCRS (Plan de Corps de Rue Simplifié – Simplified Street Plan), which is an obligation for Alteia’s end customer.
To train the algorithm, we need aerial images on which network landmarks are precisely annotated.
The annotation task then consisted of delimiting pulling chambers visible from the sky.
To carry out this annotation project, Alteia asked People for AI for assistance.
OUR SOLUTION
OUR SOLUTION
For this expert task, we set up close communication with the client via the messaging tool Slack.
We used the annotation tool recommended by the client, Labelbox. This enables the client to provide an annotation interface that is immediately ready to use and where he could add the data himself.
Labelbox also allows the creation of questions or ‘issues’ attached to the annotated images. This enables the annotation team to ask specific questions about certain images, and the customer to ask us for improvements if necessary.
All these advantages have helped the selected labelers to develop their skills more quickly.
Our team has to adapt to the specific local features (buildings, vegetation, lighting) of each new region.
We take between 3 and 5 minutes of annotation per aerial image.
We have selected and trained four labelers. The selection was based on the successful experience of previous projects using satellite images.
We have used an original method which consists of dealing with the most difficult cases by checking them on Google Street View.
Matthieu Warnier
Data Labeling Director @ People for AI
« We are currently in the fifth iteration on this task. The Alteia team and the People For AI team now know each other perfectly. This allows us to produce annotations quickly and with a high level of quality for each new request from Alteia. »
OUR IMPACT
OUR IMPACT
The accuracy of Alteia's models for detecting pulling chambers from aerial views now exceeds 95% (see Orange case study on the Alteia website)
A total of 6 man-months have been completed since the start of the annotation project.
Success after success, Alteia has become a long-term partner for People for AI.
Here are the questions our labelers had to answer to annotate this project correctly:
Manhole cover or pulling chamber?
Sewer grate or pulling chamber?
Do you think you can distinguish these different classes from the sky?
The efficiency and responsiveness of People For AI’s annotation teams had a major impact on Alteia’s ability to develop and then industrialize its Telecom network infrastructure mapping tool. The PFAI teams are continuing to support Alteia by helping to deploy the tool to new cities.
Stéphane Terrenoir
Machine Learning Project Manager @ Alteia
Client of People For AI from 2021
« People for AI is a team of efficient professionals who deliver fast and accurate annotation work. Their ability to keep the same team from one annotation campaign to the next contributes greatly to the quality of the work produced. »
Our team is now highly experienced in the annotation of pulling chambers. They can respond quickly and efficiently to any request from Alteia to adapt their algorithm to new cities.
They trust us
Our labeled data will exceed your expectations.
We provide training data. We do not provide your data to anyone else and we will not subscribe you to any newsletter without your consent.
