PhD Thesis

Deep learning-based scene understanding for autonomous vehicles

Work default illustration



  • Started: 01/01/2023
  • Thesis project read: 29/02/2024


The increasing demand for autonomous vehicles has led to a growing need for robust and reliable methods for interpreting the visual information captured by the vehicle's sensors. One key technique for achieving this goal is semantic segmentation, which involves assigning a semantic label or class to each pixel in an image. The goal of this PhD research is to develop and evaluate novel methods for semantic segmentation of images for autonomous vehicles.

The research will focus on three main areas:

* Developing new deep learning-based algorithms for semantic segmentation that are specifically tailored to the unique challenges of autonomous vehicle applications, such as handling large amounts of data, dealing with complex and dynamic scenes, and coping with limited computational resources.

* Evaluating the performance of the developed algorithms on large-scale datasets of real-world images captured by cameras and lidar sensors mounted on autonomous vehicles. This will involve using metrics such as accuracy, precision, recall, and F1-score to quantify the performance of the algorithms.

* Investigating the impact of semantic segmentation on the overall performance of autonomous vehicles by integrating the developed algorithms into a simulation environment and in real robotic hardware; and evaluating the vehicle's ability to navigate safely and efficiently.

Overall, this PhD research aims to contribute to the development of safer and more reliable autonomous vehicles by providing a better understanding of how to use semantic segmentation to interpret the visual information captured by the vehicle's sensors.