Towards Multi-Object-Tracking with Radar on a Fast Moving Vehicle: On the Potential of Processing Radar in the Frequency Domain

2020-11-26 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning · Depth: Expert, long

Summary

Researchers from Constructor University Bremen and Constructor Knowledge Labs propose processing automotive radar data in the frequency domain to enhance robustness against noise and structural errors, particularly for multi-object tracking (MOT) in high-dynamic scenarios like autonomous overtaking. This approach, exemplified by Fourier SOFT in 2D (FS2D), leverages correlation-based methods that inherently provide information about all moving structures in a scene, eliminating the need for a priori knowledge of object counts. Initial experiments using the Boreas dataset and a Navtech CIR304-H radar demonstrate radar-only odometry, achieving an average rotation error of 0.62 degrees with 1.19% outliers and a translation error of 0.49 meters, despite significant noise, motion, and Doppler distortions. This method offers a robust alternative to feature-based radar processing, which is often limited by local information susceptibility to noise.

Key takeaway

For research scientists developing autonomous driving systems, especially those focused on robust perception in challenging conditions, you should investigate frequency domain processing for radar data. This approach, demonstrated by FS2D, offers superior resilience to noise and structural errors compared to feature-based methods, and inherently simplifies multi-object tracking by identifying all dynamic elements without prior knowledge. Integrating this into your perception stack could significantly improve performance in high-dynamic scenarios like overtaking maneuvers.

Key insights

Frequency domain processing of radar data enhances robustness and inherently detects all moving objects for multi-object tracking.

Principles

• Frequency domain processing uses multi-scale information.
• Correlation methods reveal all moving structures.
• Radar odometry can be effective without sensor fusion.

Method

The Fourier SOFT in 2D (FS2D) method estimates 2D rotation via SO(3) Fourier transform projections of spectral magnitudes, performing pairwise radar scan registrations to derive motion estimates.

In practice

• Apply FS2D for robust radar-only odometry.
• Utilize frequency domain for multi-object tracking.
• Consider radar for adverse weather autonomous driving.

Topics

• Frequency Domain Processing
• Multi-Object Tracking
• Radar Odometry
• Fourier SOFT in 2D (FS2D)
• Autonomous Driving

Best for: Research Scientist, AI Scientist, Robotics Engineer, Machine Learning Engineer

Related on AIssential

• Towards Multi-Object-Tracking with Radar on a Fast Moving Vehicle: On the Potential of Processing Radar in the Frequency Domain — Processing radar data in the frequency domain improves multi-object tracking robustness in dynamic environments.
• FS-DVS: A Frequency-Selective Dynamic Visual Sensing Paradigm for Enhancing Information Completeness — FS-DVS uses a learnable pre-trigger spatial filter to mimic RGCs, enhancing event camera data completeness and noise resilience.
• RCGDet3D: Rethinking 4D Radar-Camera Fusion-based 3D Object Detection with Enhanced Radar Feature Encoding — Optimizing 4D radar feature extraction significantly boosts 3D object detection accuracy and speed, outperforming complex fusion methods.
• RISE: Single Static Radar-based Indoor Scene Understanding — RISE utilizes multipath radar reflections and generative AI for privacy-preserving, geometry-aware indoor scene understanding from a single static sensor.
• Beyond Bayer: Task-Optimal Sensor Co-Design for Robust Autonomous-Driving Segmentation — Optimizing camera sensor parameters, especially CFA weights, significantly improves autonomous driving segmentation robustness.
• Weakly Supervised Cross-Modal Learning for 4D Radar Scene Flow Estimation — Weakly supervised 4D radar scene flow estimation can achieve superior results using only image and odometry supervision.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.