Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep Radar Inverse Sensor Models for Dynamic Occupancy Grid Maps

Zihang Wei, Rujiao Yan, Matthias Schreier

TL;DR

This work proposes a deep learning-based Inverse Sensor Model (ISM) to learn the mapping from sparse RADAR detections to polar measurement grids, and is the first one to learn a single-frame measurement grid in the polar scheme from RADARs with a limited Field of View (FOV).

Abstract

To implement autonomous driving, one essential step is to model the vehicle environment based on the sensor inputs. Radars, with their well-known advantages, became a popular option to infer the occupancy state of grid cells surrounding the vehicle. To tackle data sparsity and noise of radar detections, we propose a deep learning-based Inverse Sensor Model (ISM) to learn the mapping from sparse radar detections to polar measurement grids. Improved lidar-based measurement grids are used as reference. The learned radar measurement grids, combined with radar Doppler velocity measurements, are further used to generate a Dynamic Grid Map (DGM). Experiments in real-world highway scenarios show that our approach outperforms the hand-crafted geometric ISMs. In comparison to state-of-the-art deep learning methods, our approach is the first one to learn a single-frame measurement grid in the polar scheme from radars with a limited Field Of View (FOV). The learning framework makes the learned ISM independent of the radar mounting. This enables us to flexibly use one or more radar sensors without network retraining and without requirements on 360° sensor coverage.

Deep Radar Inverse Sensor Models for Dynamic Occupancy Grid Maps

TL;DR

This work proposes a deep learning-based Inverse Sensor Model (ISM) to learn the mapping from sparse RADAR detections to polar measurement grids, and is the first one to learn a single-frame measurement grid in the polar scheme from RADARs with a limited Field of View (FOV).

Abstract

To implement autonomous driving, one essential step is to model the vehicle environment based on the sensor inputs. Radars, with their well-known advantages, became a popular option to infer the occupancy state of grid cells surrounding the vehicle. To tackle data sparsity and noise of radar detections, we propose a deep learning-based Inverse Sensor Model (ISM) to learn the mapping from sparse radar detections to polar measurement grids. Improved lidar-based measurement grids are used as reference. The learned radar measurement grids, combined with radar Doppler velocity measurements, are further used to generate a Dynamic Grid Map (DGM). Experiments in real-world highway scenarios show that our approach outperforms the hand-crafted geometric ISMs. In comparison to state-of-the-art deep learning methods, our approach is the first one to learn a single-frame measurement grid in the polar scheme from radars with a limited Field Of View (FOV). The learning framework makes the learned ISM independent of the radar mounting. This enables us to flexibly use one or more radar sensors without network retraining and without requirements on 360° sensor coverage.
Paper Structure (23 sections, 6 equations, 8 figures, 4 tables)

This paper contains 23 sections, 6 equations, 8 figures, 4 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. Geometric ISMs
  4. Deep ISMs
  5. Dynamic Grid Maps (DGMs)
  6. Method
  7. Evidential Occupancy Representation
  8. Generation of RADAR Input Data
  9. Generation of LiDAR-based Labels
  10. Model
  11. Network architecture
  12. Loss
  13. EXPERIMENTAL SETUP
  14. Baseline: RADAR Geometric ISM
  15. Data Collection
  16. ...and 8 more sections

Figures (8)

  • Figure 1: Overview of geometric ISMs (top) and deep ISMs (bottom). For illustrative purposes, the 2D RADAR point cloud is colored in pink. A front RADAR is chosen to show the exemplary Measurement Grid (MG). The MGs are then used as inputs for a dynamic grid fusion algorithm to estimate a DGM, which jointly represents the static and dynamic driving environment in a consistent manner. The ego-vehicle is denoted as a black dot with a directional bar in the center of the DGMs, free areas are shown in green, statically occupied areas in red, and dynamically occupied areas in blue.
  • Figure 2: Exemplary illustration of the involved coordinate conversion on the front RADAR in a highway scenario. a) Front camera image. b) $\textbf{g}^{(evi)}$ of the area related to the front RADAR in Cartesian coordinates. c) $\textbf{g}^{(evi)}$ in polar coordinates. A guardrail (red) and a vehicle (blue) are exemplarily shown as corresponding ellipses in all images. $r$: range, $\phi$: azimuth angle.
  • Figure 3: Driven routes for dataset generation: test set (red), training and validation set (black)
  • Figure 4: Qualitative result on setup B with different RADARs, here a) front center RADAR, b) front left RADAR, c) rear center RADAR. Rows from top to bottom are: filtered LiDAR-based annotation before thresholding, RADAR detections, RADAR measurement grid. The dynamic object is indicated manually with a bounding box.
  • Figure 5: Polar measurement grid comparison between single frame and double frame. Here different RADARs are shown: a) front center, single frame. b) front center, double frames. c) front left, single frame. d) front left, double frames. The dynamic object is indicated with a bounding box. Rows from top to bottom are: filtered LiDAR-based annotation before thresholding, RADAR detections, RADAR measurement grid.
  • ...and 3 more figures