# Deep Learning on Radar Centric 3D Object Detection

**Authors:** Seungjun Lee  
**arXiv:** [2003.00851](https://arxiv.org/abs/2003.00851)  
**Date:** 2020-02-27  
**Summary page:** https://www.summarizepaper.com/en/paper/2003.00851v1/

## Key points

- Seungjun Lee introduces a novel approach to 3D object detection using radar data
- Existing algorithms heavily rely on camera and LiDAR sensors, which are susceptible to harsh weather and lighting conditions
- Radar is resistant to such conditions, but deep learning on radar data has only recently been explored
- The authors present a deep learning-based model for 3D object detection using radar data exclusively
- They are the first to demonstrate the effectiveness of a deep learning approach trained on a public radar dataset
- Limited labeled radar data is overcome by leveraging abundant LiDAR data through transformation techniques and aggressive radar augmentation methods
- The proposed model achieves accurate and robust 3D object detection in challenging weather and lighting conditions
- Radar shows potential as a reliable sensor for object detection tasks
- This research opens up new possibilities for enhancing autonomous systems' perception capabilities in adverse environmental conditions where traditional sensors may struggle.

## Summary

In this paper titled "Deep Learning on Radar Centric 3D Object Detection," Seungjun Lee introduces a novel approach to 3D object detection using radar data. While existing algorithms heavily rely on camera and LiDAR sensors, these are susceptible to harsh weather and lighting conditions. In contrast, radar is resistant to such conditions. However, the application of deep neural networks on radar data has only recently been explored. The authors present a deep learning-based model for 3D object detection that solely utilizes radar data. This is a significant contribution as they are the first to demonstrate the effectiveness of a deep learning approach trained on a public radar dataset. To overcome the challenge of limited labeled radar data, the authors propose an innovative solution by leveraging abundant LiDAR data through transformation techniques and employing aggressive radar augmentation methods. By combining these approaches, the proposed model achieves accurate and robust 3D object detection in challenging weather and lighting conditions. The results highlight the potential of utilizing radar as a reliable sensor for object detection tasks. This research opens up new possibilities for enhancing autonomous systems' perception capabilities, particularly in adverse environmental conditions where traditional sensors may struggle. Overall, this paper contributes to advancing the field of 3D object detection by introducing a deep learning-based approach that leverages exclusively radar data while also making use of abundant LiDAR data through transformation techniques to further enhance its performance in detecting objects accurately under challenging conditions.