A New Automotive Radar 4D Point Clouds Detector by Using Deep Learning

Abstract

The millimeter-wave radar, as an important sensor, is widely used in autonomous driving. In recent years, to meet the requirement of high level autonomous driving applications, attentions have been paid to generate high-quality radar point clouds. However, in the complex roadway environment, the weaknesses of classical radar detectors are exposed, such as too much clutter points and sparse valid point clouds. Therefore, in this paper, we propose a new automotive radar detector based on deep learning using the spatial distribution feature of the real targets, in order to improve the performance of automotive radar detector in the real-world driving scene. Besides, aiming at the lack of radar data labels, we propose an autonomous labeling method by using synchronized Lidar data. Finally, we evaluate the detector on data collected in the real-world roadway scene and the result shows that the proposed radar detector out-performs the classical radar detectors in suppressing the clutter and generating denser point clouds.

High Fidelity Physics-Based Simulation of a 512-Channel 4D-Radar Sensor for Automotive Applications

“…The range-Doppler maps (RDM) were grouped for azimuth and elevation AoA determination. A RDM from a single channel was then fed to a cell-averaging (CA) constant false alarm rate detector (CFAR) [24] to produce a hit matrix. This hit matrix mask was then applied to the channel RDMs for azimuth and elevation angle of arrival (AoA) determination.…”

“…), maneuvered around or avoided by braking in the worst case as part of active-safety definition [21]. Recent efforts have addressed this need by introducing MIMO 4D radar sensors that can determine the range, velocity, azimuth and elevation AoA of a target and produce a point cloud from the traffic scene [7], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30]. A sensor-fusion strategy involving a 4D radar sensor and monovision camera was used in [22] to conduct 3D mapping of traffic scenes.…”

Towards Deep Radar Perception for Autonomous Driving: Datasets, Methods, and Challenges

“…Neural networks are expected to better utilise this information. One option is to use neural networks to replace CFAR [156] or DOA estimation [76,157]. Readers can refer to [158] for a detailed survey of learning-based DOA estimation.…”

Recent Advancements in Learning Algorithms for Point Clouds: An Updated Overview

“…FMCW radar has the peculiarity that it can change its operating frequency during the measurement, i.e., the transmission signal is modulated in frequency (or in phase). FMCW radars offer robust sensing to autonomous vehicles [ 27 , 36 ], for their high-range resolution and accuracy: in fact, FMCW radars add an extra dimension in the sensing, and they are more robust to weather changes, with respect to LiDAR sensors. FMCW radars are often employed also in Human Motion Detection [ 25 , 37 ] and Activity Recognition systems [ 28 ].…”

“…This allows one to canonicalize point order and learn generalized convolutional features from unordered and unstructured point clouds. On the other hand, the introduction of skip connections, as in [ 36 ], has proven to boost the performance of convolutional networks: a UNet-like architecture is adopted in this work for classification and object detection in mmWave-radar point clouds.…”