[论文阅读] DAY 1 Vehicle Detection With Automotive Radar Using Deep Learning on Range-Azimuth-Doppler

Vehicle Detection With Automotive Radar Using Deep Learning on Range-Azimuth-Doppler Tensors

作者: Qualcomm AI Research∗ Qualcomm Technologies, Inc.†

原文: 原文地址链接

Abstract

二十多年来，雷达一直是汽车高级驾驶员辅助系统的关键推动者。作为一种廉价、全天候和远程的同时提供速度测量的传感器，雷达被认为是未来自主驾驶不可或缺的。传统的雷达信号处理技术往往不能将目标的反射和杂波区分开来，通常仅限于检测接收信号中的峰值。这些峰值检测方法有效地将像图像一样的雷达信号压缩成稀疏的点云。本文提出了一种基于深度学习的车辆检测方案，该方案基于图像的张量代替峰值检测产生的点云。

Radar has been a key enabler of advanced driver assistance systems in automotive for over two decades.
Being an inexpensive, all-weather and long-range sensor that simultaneously provides velocity measurements, radar is expected to be indispensable to the future of autonomous driving. Traditional radar signal processing techniques often cannot distinguish reflections from objects of interest from clutter and are generally limited to detecting peaks in the received signal.
These peak detection methods effectively collapse the image-like radar signal into a sparse point cloud. In this paper, we demonstrate a deep-learning-based vehicle detection solution which operates on the image-like tensor instead of the point cloud resulted by peak detection.
To the best of our knowledge, we are the first to implement such a system.

论点

• Radar 是3D lidar 的一种低成本替代方案
• Radar 对光线的变化 以及 雨和雾天气的表现更加稳定, 相比于 Camera 和 Lidar
• 由于Radar 是基于测量电磁反射的强度来判断障碍物, 所以如何合理切直观的表示更加困难
• 通常的处理方法, 将原始数据(3D tensor, 和多普勒效应产生的速度值)转换成稀疏的2D 点云地图, 将障碍物和周围的环境分隔开来, 这样会损失很多的信息.
• 本文提供了: 直接作用于 3D tensor 的神经网络.
  • 数据处理利用 FFT 将原始数据转换成 频域的三维 Tensor, 然后送入如下的 DP模型.
    

结果