The Radar Ghost Dataset -- An Evaluation of Ghost Objects in Automotive Radar Data

TL;DR

This paper tackles the persistent issue of ghost objects in automotive radar by introducing a richly annotated dataset with 111 sequences across 21 scenarios, detailing multiple ghost types (type-1, type-2, and higher-order bounces) and providing synthesized overlays to enable multi-object experimentation. It presents two baseline processing pipelines—semantic segmentation with PointNet++ and instance segmentation with SGPN—along with adaptations to handle radar data, ignore regions, and background clutter. Comprehensive experiments show that Doppler-based features are crucial for ghost discrimination, that SGPN excels at ghost-type separation while PointNet++ better suppresses false positives, and that ghost objects can meaningfully inflate FP rates in detection pipelines. By enabling detailed ghost-object analysis and offering synthesized multi-object data, the work aims to advance data-driven radar multi-path suppression or exploitation for robust vehicle perception.

Abstract

Radar sensors have a long tradition in advanced driver assistance systems (ADAS) and also play a major role in current concepts for autonomous vehicles. Their importance is reasoned by their high robustness against meteorological effects, such as rain, snow, or fog, and the radar's ability to measure relative radial velocity differences via the Doppler effect. The cause for these advantages, namely the large wavelength, is also one of the drawbacks of radar sensors. Compared to camera or lidar sensor, a lot more surfaces in a typical traffic scenario appear flat relative to the radar's emitted signal. This results in multi-path reflections or so called ghost detections in the radar signal. Ghost objects pose a major source for potential false positive detections in a vehicle's perception pipeline. Therefore, it is important to be able to segregate multi-path reflections from direct ones. In this article, we present a dataset with detailed manual annotations for different kinds of ghost detections. Moreover, two different approaches for identifying these kinds of objects are evaluated. We hope that our dataset encourages more researchers to engage in the fields of multi-path object suppression or exploitation.

Figures (7)

• Figure 1: Our dataset provides detailed annotations for various types of ghost targets allowing to conduct in-depth experiments on their influence. Example image from our dataset. Different types of multi-path reflections are highlighted alongside the real object. The small white dots correspond to a lidar reference system, the gray ones are the radar reflections. See \ref{['sec:radar-ghosts']} for a detailed explanation of radar ghost objects.
• Figure 2: Recorded scenario and reflective surface examples.
• Figure 3: Radar sensor specification and sensor mounting positions schematic. Upper: Frequency band $f$, range $r$, azimuth angle $\phi$, and radial (Doppler) velocity $v_r$. Lower: Resolutions $\Delta$ for $r$, $\phi$, $v_r$, and time $t$. Adapted from Scheiner2020CVPR and Kraus2020ITSC.
• Figure 4: Overview of the labeling process. The red lines on the radar detections are a visualization of the Doppler velocity, longer lines correspond to higher Doppler values. Cf. \ref{['tab:annotations']} for abbreviations. Best viewed on screen.
• Figure 5: Results for pedestrian vs. cyclist. Average precision (AP) per class and average (mAP). Scores are given at two different IoU thresholds.