DHI

From annotated data to automatic bird tracking

Move has developed an AI-based video tracker that automatically finds and follows birds and delivers recordings to DHI's MUSE system for species identification. The work has covered data annotation, model training, camera control and ongoing optimisation during roll-out. Today we also manufacture the finished tracker servers in our own production.

Customer
DHI
Product
AI-based video tracker for the MUSE system
Move's role
Data annotation, AI model development, real-time tracking, camera control, system integration, testing and production
Mathieu Bulliot

Move Innovation feels like an extension of our own team rather than a traditional supplier. They base decisions on data, analysis and process, which sets them apart from other consultants, who are often inexperienced and wing it.

Move know their craft and have the right skills, which gives us confidence in the collaboration. At the same time, they are approachable and easy to talk to, so we can have an open dialogue. That makes it easy to find good solutions together.

Mathieu Bulliot

Engineer, DHI

Michael Hass

They are not a big, anonymous business, and they focus on collaboration rather than a transactional approach.

We value the team's broad skills, the realistic estimates and the honest dialogue about budget, timeline and any limitations – even when there is something they have not tried before.

Michael Hass

Decision maker, DHI

Bird monitoring at offshore wind farms

DHI has developed MUSE to monitor birds and their movements at wind farms. The system combines sensors and data analysis, while MUSE AI adds automatic species identification.

The information can be used to support the protection of selected bird species, for example by slowing down or stopping individual wind turbines when there is a risk of collision.

Move's video tracker contributes the recordings that are passed on to species identification in MUSE.

From a bird in flight to usable video

DHI needed a solution that could automatically find and follow birds in real time.

The birds can be small in the image, move fast and be far from the camera. The task therefore required close interplay between the AI model, the ongoing tracking and the camera's movements, so the bird could be kept in the recording.

Annotated data as the basis for the AI model

A central part of the development was creating the data basis for model training.

Image material was annotated so that the birds were marked in the data the model was to learn from. Annotation and training were thus part of the work of developing a model that could find birds in the camera image.

Model development was closely linked to the overall solution: the detections had to be usable for controlling the camera and creating usable video sequences.

Ongoing optimisation during roll-out

The AI model was developed further as the solution was rolled out.

Recordings and experience from practical use provided a basis for assessing the model's behaviour and continuing work on the data, annotation and training. Development thus continued through an ongoing process in which experience from the installations was fed back into the model.

This made it possible to improve the solution based on the situations it actually encountered.

AI and camera control in real time

The tracker server controls a PTZ camera that can pan, tilt and zoom. The solution finds the bird, zooms in and follows its movements automatically while the video sequence is recorded.

AI processing takes place locally on the server, and the solution supports several camera systems. It is designed for unmanned operation at remote locations, with remote monitoring and support.

From field testing to production

Development and testing were carried out in close collaboration with DHI. Through workshops and joint field tests, DHI's team was also equipped to install and operate the systems themselves.

After the development phase, the hardware was optimised for scaling. The tracker servers are now manufactured in Move's own production.

The result: an AI solution that develops through practical use

DHI now has automatic bird tracking that delivers video for further analysis in the MUSE system.

The project brings together data work, model development, software, camera control and hardware production. At the same time, ongoing optimisation has made experience from the roll-out an active part of the solution's further development.

Technologies and skills

AI and data

  • Data annotation and building the training data
  • Training and ongoing optimisation of AI models
  • Object detection and tracking in real time

Camera and integration

  • Automatic PTZ camera control and edge AI
  • Integration with DHI's MUSE system

Operations and production

  • Field testing, remote monitoring and support
  • Production readiness and manufacture of tracker servers
Ready when you are

Does your AI solution need to work outside the lab?

We help with the whole journey from data and model training to software, hardware and production – with ongoing improvements based on experience from practical use.

Reach out to us →