facebook tracking

Master Thesis - 3D Lane Detection at Edge

Background 

The performance of many autonomous driving features such as lane-keeping assist, lane departure warning, and pilot assist is directly dependent on the performance of the lane detection module.

The task of lane detection has lately been dominated by deep learning approaches for monocular vision, which aim to predict lanes in the 2D image plane. These methods usually consider flat earth assumption or follow a multi-view geometry approach with constant lane width assumption to project the detected 2D lanes to the 3D world. Recently, lane detection has been extended to end-to-end deep learning models that instead predict 3D lanes directly in the world. These models overcome some of the shortcomings of the previous methods, including restrictions in lane structures and the difficulties involved in lane model fitting and image-to-world correspondence.

These models still need to be trained on a massive amount of data collected by cars driving worldwide to deliver the target performance. But data collection and data transfer are not straightforward due to the emerging data-sharing policies. So, we need a solution that keeps data where it resides and instead brings AI models to the data. This solution is called Federated Learning (FL). FL is an emerging technology that enables training AI models across multiple decentralized servers or edge devices like autonomous cars while ensuring data privacy and security by design. It removes the need to transfer vast amounts of data out of the vehicles, resulting in optimized bandwidth and storage use while the acquired insights from data remain intact. It also enables shorter improvement cycles by utilizing more diverse data collected from various edge devices.

However, training 3D lane detection models in a federated learning setup requires accessing superior annotations in the car, which is challenging not having access to the skilled human annotators at the edge nor having costly and high-precision sensors. Therefore, it is crucial to have some automatic mechanism to annotate data with high enough quality to enable training 3D lane detection models in the car. 

Project Description 

In this master thesis project, you will focus on: 

  • Automatic annotation methods to generate 3D lane annotations in the car. Data curation will be provided for selecting candidate frames for annotation. Automatic annotation methods will apply to the real-world dataset from Zenseact.
  • Improve the semi-supervised 3D lane detection method developed at Zenseact.
  • Train the semi-supervised 3D lane detection model with the automatically generated 3D lane annotations and the available unlabeled data at the edge.
  • Deploy the federated 3D lane detection in simulation mode initially and finally at the edge in federated mode.
  • Demonstrate results and present findings at Zenseact. 

Qualifications 

We are looking for two highly motivated students with a good general background in machine learning and computer vision. The following skills would be precious:

  • Deep learning
  • Federated learning
  • Python programming
  • Handling complex systems  

Further information

Please send in individual applications with CV, motivational letter, and grade transcripts. 

Planned start: January 2022, with some flexibility.

Final application date: 15 November 2021, but we will screen candidates continuously, so please submit your application as soon as possible.

Duration: 30 ECTS 

For questions regarding the project, please contact: mina.alibeigi@zenseact.com, or benny.nilsson@zenseact.com.

Additional information

  • Remote status

    Flexible remote

Or, know someone who would be a perfect fit? Let them know!

Gothenburg, Sweden

Lindholmspiren 2
417 56 Göteborg Directions View page

Making safe and intelligent mobility real.

At Zenseact, we lead the global movement of crafting tomorrow's mobility with the software platform of choice. Our mission is to “Make safe and intelligent mobility real, for everyone, everywhere”. This statement marks our conviction and dedication to bring autonomous driving out on the streets for real and is at the center of everything we do.

We could not dream of achieving this without our great teams of very talented people. We are on this journey together and our agile way of working is reflected throughout our entire organization; it is part of our culture and how we work, develop and grow together.

Teamtailor

Applicant tracking system by Teamtailor