Cao and Sengupta Automate Radar-Camera Dataset Labeling with AI

Dr. Cao and grad student Arindam Sengupta propose a method to auto-label radar data using object detection from camera streams.

In March 2022, Dr. Siyang Cao along with graduate student Arindam Sengupta published a new method to automatically generate labeled radar data aligned with camera images, by leveraging object detection (YOLO) on images and associating radar point clouds accordingly.

Addressing a Major Bottleneck

This work addresses a critical challenge in deep learning for radar: the lack of annotated radar datasets. Manual labeling of radar data is time-consuming and requires specialized expertise.

Impact and Applications

Their automated pipeline:

  • Drastically reduces manual labeling effort
  • Enhances training of radar-only perception models
  • Accelerates research in autonomous sensing
  • Benefits healthcare monitoring applications
  • Improves transportation domain research

The lab highlighted that this tool can serve as a foundation for advancing radar imaging and perception across multiple domains.

Published on March 05, 2022