Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PillarGen: Enhancing Radar Point Cloud Density and Quality via Pillar-based Point Generation Network

Jisong Kim, Geonho Bang, Kwangjin Choi, Minjae Seong, Jaechang Yoo, Eunjong Pyo, Jun Won Choi

TL;DR

The experiments demonstrate that PillarGen outperforms traditional point upsampling methods in quantitative and qualitative measures and it is confirmed that when PillarGen is incorporated into bird’s eye view object detection, a significant improvement in detection accuracy is achieved.

Abstract

In this paper, we present a novel point generation model, referred to as Pillar-based Point Generation Network (PillarGen), which facilitates the transformation of point clouds from one domain into another. PillarGen can produce synthetic point clouds with enhanced density and quality based on the provided input point clouds. The PillarGen model performs the following three steps: 1) pillar encoding, 2) Occupied Pillar Prediction (OPP), and 3) Pillar to Point Generation (PPG). The input point clouds are encoded using a pillar grid structure to generate pillar features. Then, OPP determines the active pillars used for point generation and predicts the center of points and the number of points to be generated for each active pillar. PPG generates the synthetic points for each active pillar based on the information provided by OPP. We evaluate the performance of PillarGen using our proprietary radar dataset, focusing on enhancing the density and quality of short-range radar data using the long-range radar data as supervision. Our experiments demonstrate that PillarGen outperforms traditional point upsampling methods in quantitative and qualitative measures. We also confirm that when PillarGen is incorporated into bird's eye view object detection, a significant improvement in detection accuracy is achieved.

PillarGen: Enhancing Radar Point Cloud Density and Quality via Pillar-based Point Generation Network

TL;DR

The experiments demonstrate that PillarGen outperforms traditional point upsampling methods in quantitative and qualitative measures and it is confirmed that when PillarGen is incorporated into bird’s eye view object detection, a significant improvement in detection accuracy is achieved.

Abstract

In this paper, we present a novel point generation model, referred to as Pillar-based Point Generation Network (PillarGen), which facilitates the transformation of point clouds from one domain into another. PillarGen can produce synthetic point clouds with enhanced density and quality based on the provided input point clouds. The PillarGen model performs the following three steps: 1) pillar encoding, 2) Occupied Pillar Prediction (OPP), and 3) Pillar to Point Generation (PPG). The input point clouds are encoded using a pillar grid structure to generate pillar features. Then, OPP determines the active pillars used for point generation and predicts the center of points and the number of points to be generated for each active pillar. PPG generates the synthetic points for each active pillar based on the information provided by OPP. We evaluate the performance of PillarGen using our proprietary radar dataset, focusing on enhancing the density and quality of short-range radar data using the long-range radar data as supervision. Our experiments demonstrate that PillarGen outperforms traditional point upsampling methods in quantitative and qualitative measures. We also confirm that when PillarGen is incorporated into bird's eye view object detection, a significant improvement in detection accuracy is achieved.
Paper Structure (21 sections, 13 equations, 3 figures, 4 tables)

This paper contains 21 sections, 13 equations, 3 figures, 4 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. Proposed PillarGen
  4. Pillar Encoding
  5. Occupied Pillar Prediction (OPP)
  6. Pillar to Point Generation (PPG)
  7. Loss function
  8. EXPERIMENTS
  9. Experimental Setup
  10. Datasets
  11. Evaluation metrics
  12. Implementation Details
  13. Baselines
  14. Quantitative Results
  15. Qualitative Results
  16. ...and 6 more sections

Figures (3)

  • Figure 1: Radar Point Cloud Data Comparison. Left: Sample from short-range radar. Right: Sample from long-range radar. The color of the points represents the radar cross-section (RCS) values, with lower values closer to blue and higher values closer to red, and the green boxes indicate the ground truth boxes.
  • Figure 2: Overall architecture of the proposed PillarGen: PillarGen takes short-range radar points as input to generate long-range radar points. The Pillar Target Generator forms GT Radar pillars using GT radar points. Pillar Encoding and 2D CNN modules extract BEV features from input radar points. OPP module uses BEV features and GT Radar pillars to simultaneously conduct three predictions. Finally, PPG module generates long-range radar points through pillar feature expansion and point generation.
  • Figure 3: Qualitative Results on our test dataset. We compare the point clouds generated from short-range radar points using different method (PU-NET, PU-GCN, Dis-PU, and ours) against long-range radar points. The color of the points represents the RCS values, with lower values closer to blue and higher values closer to red and the green boxes indicate the ground truth boxes.