Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BoostRad: Enhancing Object Detection by Boosting Radar Reflections

Yuval Haitman, Oded Bialer

TL;DR

The paper tackles the challenge of wide angular PSF in automotive radar that causes blur and clutter by proposing BoostRad, a two-stage system where a Reflection Boosting Network sharpens the radar image (narrowing the PSF) and a subsequent Object Detection Network detects objects on the boosted image. Training leverages a physics-informed ground-truth probability map derived from a high-resolution 'super-radar' reference, with synthetic data generated via a CARLA-based radar simulator to overcome the lack of narrow-PSF hardware. Experiments on RADDet and CARRADA show BoostRad outperforms end-to-end baselines and ablation studies highlight the importance of the PSF-focused loss components and the chosen resolution enhancement factor $\kappa$ (optimal near 12). The approach demonstrates strong practical impact and suggests broader applicability to sensors with wide PSFs, while challenging the dominance of single-stage end-to-end methods in radar perception and enabling synthetic-data-driven training.

Abstract

Automotive radars have an important role in autonomous driving systems. The main challenge in automotive radar detection is the radar's wide point spread function (PSF) in the angular domain that causes blurriness and clutter in the radar image. Numerous studies suggest employing an 'end-to-end' learning strategy using a Deep Neural Network (DNN) to directly detect objects from radar images. This approach implicitly addresses the PSF's impact on objects of interest. In this paper, we propose an alternative approach, which we term "Boosting Radar Reflections" (BoostRad). In BoostRad, a first DNN is trained to narrow the PSF for all the reflection points in the scene. The output of the first DNN is a boosted reflection image with higher resolution and reduced clutter, resulting in a sharper and cleaner image. Subsequently, a second DNN is employed to detect objects within the boosted reflection image. We develop a novel method for training the boosting DNN that incorporates domain knowledge of radar's PSF characteristics. BoostRad's performance is evaluated using the RADDet and CARRADA datasets, revealing its superiority over reference methods.

BoostRad: Enhancing Object Detection by Boosting Radar Reflections

TL;DR

The paper tackles the challenge of wide angular PSF in automotive radar that causes blur and clutter by proposing BoostRad, a two-stage system where a Reflection Boosting Network sharpens the radar image (narrowing the PSF) and a subsequent Object Detection Network detects objects on the boosted image. Training leverages a physics-informed ground-truth probability map derived from a high-resolution 'super-radar' reference, with synthetic data generated via a CARLA-based radar simulator to overcome the lack of narrow-PSF hardware. Experiments on RADDet and CARRADA show BoostRad outperforms end-to-end baselines and ablation studies highlight the importance of the PSF-focused loss components and the chosen resolution enhancement factor (optimal near 12). The approach demonstrates strong practical impact and suggests broader applicability to sensors with wide PSFs, while challenging the dominance of single-stage end-to-end methods in radar perception and enabling synthetic-data-driven training.

Abstract

Automotive radars have an important role in autonomous driving systems. The main challenge in automotive radar detection is the radar's wide point spread function (PSF) in the angular domain that causes blurriness and clutter in the radar image. Numerous studies suggest employing an 'end-to-end' learning strategy using a Deep Neural Network (DNN) to directly detect objects from radar images. This approach implicitly addresses the PSF's impact on objects of interest. In this paper, we propose an alternative approach, which we term "Boosting Radar Reflections" (BoostRad). In BoostRad, a first DNN is trained to narrow the PSF for all the reflection points in the scene. The output of the first DNN is a boosted reflection image with higher resolution and reduced clutter, resulting in a sharper and cleaner image. Subsequently, a second DNN is employed to detect objects within the boosted reflection image. We develop a novel method for training the boosting DNN that incorporates domain knowledge of radar's PSF characteristics. BoostRad's performance is evaluated using the RADDet and CARRADA datasets, revealing its superiority over reference methods.
Paper Structure (21 sections, 9 equations, 9 figures, 6 tables)

This paper contains 21 sections, 9 equations, 9 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Radar Object Detection
  4. Radar Datasets
  5. Object Detection With Reflection Boosting
  6. Reflection Boosting Network
  7. Object Detection Network
  8. Synthetic Data Generation
  9. Experiments and Results
  10. Object Detection Performance Evaluation
  11. Qualitative Performance Evaluation
  12. Ablation Study
  13. Conclusion
  14. Supplementary Performance Evaluation
  15. Reflection Boosting Performance Evaluation
  16. ...and 6 more sections

Figures (9)

  • Figure 1: Radar's point spread function (PSF).
  • Figure 2: Comparison of low-resolution realistic automotive radar reflection image (a) and high-resolution 'super-radar' image (b). White points depict vehicle, black asterisk shows radar reflection point from vehicle. Reflection points in (a) have wide PSF, while in (b) the PSF is narrow.
  • Figure 3: BoostRad Overview: The boosting DNN enhances input reflection image $\mathcal{H}_{input}$ to a radar image $\mathcal{H}_{boost}$ with a narrower PSF, resembling the 'super-radar' image $\mathcal{H}_{super}$. Subsequently, object detection DNN identifies objects in the boosted image.
  • Figure 4: Diagram depicting reflection boosting DNN architecture with 4x expansion in output angular dimension compared to input.
  • Figure 5: Radar simulation pipeline with illustrated simulated scenario: camera image, reflection points, and radar reflection image.
  • ...and 4 more figures