Description
Radar sensors are increasingly valued in autonomous systems due to their low cost, long-range capability, and resilience to adverse weather conditions. However, the sparsity and noise inherent in radar point clouds limit their effectiveness for tasks such as object detection and scene understanding. In contrast, LiDAR sensors produce dense and accurate data, but they come with higher costs and are sensitive to environmental factors. This thesis explores the use of diffusion models to enhance sparse radar point clouds by generating denser and potentially more informative representations. By learning the underlying data distribution through iterative denoising, diffusion-based methods offer a promising direction for improving the usability of radar data. The goal is to investigate whether such models can bridge the performance gap between radar and denser sensing modalities like LiDAR, enabling more robust and cost-effective perception systems.