Exploring Radar Data Representations in Autonomous Driving: A Comprehensive Review

TL;DR

This comprehensive review addresses the problem of robust radar perception for autonomous driving by analyzing diverse radar data representations. It introduces five representations—ADC signal, radar tensor, point cloud, grid map, and micro-Doppler signature—and discusses how each maps raw measurements to task-relevant features. The authors synthesize datasets and processing methods across 2019–2024, identify key advantages and limitations, and propose directions such as 4D radar, BEV-based fusion, and transformer-based architectures. The work provides guidance for radar perception researchers and offers an interactive website to facilitate dataset and method comparison.

Abstract

With the rapid advancements of sensor technology and deep learning, autonomous driving systems are providing safe and efficient access to intelligent vehicles as well as intelligent transportation. Among these equipped sensors, the radar sensor plays a crucial role in providing robust perception information in diverse environmental conditions. This review focuses on exploring different radar data representations utilized in autonomous driving systems. Firstly, we introduce the capabilities and limitations of the radar sensor by examining the working principles of radar perception and signal processing of radar measurements. Then, we delve into the generation process of five radar representations, including the ADC signal, radar tensor, point cloud, grid map, and micro-Doppler signature. For each radar representation, we examine the related datasets, methods, advantages and limitations. Furthermore, we discuss the challenges faced in these data representations and propose potential research directions. Above all, this comprehensive review offers an in-depth insight into how these representations enhance autonomous system capabilities, providing guidance for radar perception researchers. To facilitate retrieval and comparison of different data representations, datasets and methods, we provide an interactive website at https://radar-camera-fusion.github.io/radar.

Figures (10)

• Figure 1: Radar perception in autonomous driving. (a) Radars on the vehicle are employed to detect objects in the path and surroundings. (b) Radars inside the cabin are leveraged to monitor occupant vital signs and behaviors. (c) Radars alongside roadways are utilized to measure the speed of passing vehicles and estimate traffic flow.
• Figure 2: Overview of this review. Section \ref{['sec:Radar Perception']} provides an overview of radar perception, including its working principles and signal processing techniques. Section \ref{['sec:Radar Data Representations']} presents an in-depth examination on datasets and methods of different radar data representations. Section \ref{['sec:Discussion']} discusses the challenges and potential directions for research and development for radar perception in autonomous driving.
• Figure 3: Overview of radar working pipeline.
• Figure 4: Pipeline of radar signal processing. (a) Frequency of chirps emitted by a TX antenna and received by an RX antenna. (b) Frequency of chirps received by multiple RX antennas. (c) Sampling performed on each chirp. (d) Sample-Chirp map generated from sampling on all chirps. (e) Simple-Chirp-Antenna tensor generated from Sample-Chirp maps based on multiple RX antennas.
• Figure 5: Overview of CFAR processing. Set of $X_1, X_2, ..., X_N$ represents the detection window, $Y$ is the value of CUT, $Z$ represents the clutter background level of the CUT, $T$ is the detection threshold, $\alpha$ donates a scaling factor, $H_1$ declares that an object is located within the CUT, $H_0$ indicates that there is no object in the current CUT.