Automotive Elevation Mapping with Interferometric Synthetic Aperture Radar

TL;DR

The paper addresses the limited elevation information in automotive radar by leveraging Interferometric SAR to recover height from phase differences across baselines. It introduces a practical signal model and a compact, vehicle‑mounted dual‑array system that forms SAR images and extracts elevation via vertical baselines, producing 3D point clouds using spherical to Cartesian deprojection with $s_x = r \cos\theta$, $s_y = r \sin\theta\cos\phi$, $s_z = r \sin\theta\sin\phi$. A low‑compute processing pipeline is demonstrated in agricultural and urban driving scenarios, achieving real‑time capable performance (sub‑second per frame) and centimeter‑level accuracy in controlled tests. The work suggests a viable path to using low‑cost automotive radar as a primary 3D perception sensor, enabling elevation‑aware obstacle understanding and drivable area occupancy for autonomous driving, with clear avenues for enhancements like higher elevation resolution and multi‑source deconvolution.

Abstract

Radar is a low-cost and ubiquitous automotive sensor, but is limited by array resolution and sensitivity when performing direction of arrival analysis. Synthetic Aperture Radar (SAR) is a class of techniques to improve azimuth resolution and sensitivity for radar. Interferometric SAR (InSAR) can be used to extract elevation from the variations in phase measurements in SAR images. Utilizing InSAR we show that a typical, low-resolution radar array mounted on a vehicle can be used to accurately localize detections in 3D space for both urban and agricultural environments. We generate point clouds in each environment by combining InSAR with a signal processing scheme tailored to automotive driving. This low-compute approach allows radar to be used as a primary sensor to map fine details in complex driving environments, and be used to make autonomous perception decisions.

Figures (6)

• Figure 1: Vertical Baseline Geometry: Illustration of the signal receive path from the source to the vertical baseline on the left. Dashed lines indicate the wavefront at regular intervals along the plane wave trajectory. The plane wave arrives at $\text{VX}_1$ with time delay $\tau_{\phi}$ relative to $\text{VX}_0$.
• Figure 2: Projection Effect: A 3D source at Cartesian coordinates $(s_x, s_y, s_z)$ with range $r$, azimuth $\theta$ and elevation $\phi$ is projected to position $(s_u, s_v)$ in range-azimuth $(u, v)$ space. As elevation increases, the projected position is further out than its 2D Cartesian position $(s_x, s_y)$.
• Figure 3: Radar Array: TI AWR1243BOOST array consists of 4 (blue squares) and 3 (orange diamonds) antenna elements. Antenna elements form a dense, 2-elevation layer virtual array of (green circles).
• Figure 4: Reflector Accuracy: a) log magnitude image of test scene with two reflectors. b) 2D projection map with points colored by elevation in meters. One reflector is placed on the ground and the other is placed on a mount at either 33 cm or 63 cm. This projection map corresponds to the 63 cm case. Other objects in the scene include an absorber structure around the reflectors, tube lights, a desk chair, a window, and a computer monitor, each labeled with a colored box.
• Figure 5: Vineyard Scene: a) Camera view containing features of interest including rows of grapevines, trees, and a person. b) SAR log magnitude image. c) 2D projection map with points colored by elevation in meters and boxes around regions of interest. d) Point cloud renderings of regions of interest. Points colored by elevation according to the color scale in c). The left purple box contains a cluster of trees with visible tree trunks, appearing above the rows of grapevines. The middle blue box contains a person kneeling down next to the grapevines. The right orange box contains rows of grapevines.