摘要
近20年来,随着以物联网技术,计算机视觉技术为代表的核心技术蓬勃发展,基于深度学习的目标检测算法在各个领域都受到了较高的重视,而车辆目标检测是基于深度学习的目标检测中的一个重要研究领域,也是应用在智能驾驶、智能交通系统中非常重要的一部分。针对车辆目标检测任务,首先对深度学习的车辆目标检测进一步探讨,提出检测任务的重点、难点及发展现状,以时间线对卷积神经网络下车辆目标检测算法进行概括,并对目前2种主流的基于候选框和基于回归的车辆目标检测算法进行总结。伴随着目标检测算法的更加轻量化,检测性能更加优越,将在嵌入式设备得到应用,以提高检测任务的效率。在未来自动驾驶,智能交通系统领域对于安全性,实时性的要求会更高,使得车辆目标检测算法有较好的发展前景。
In the past 20 years,with the vigorous development of core technologies represented by Internet of things technology and computer vision technology,object detection algorithms based on deep learning have received high attention in various fields,and vehicle object detection is an important research field of object detection based on deep learning.It is also a very important part of intelligent driving and intelligent transportation system.For vehicle detection task,first of all,the vehicle object in the deep learning is further discussed,the emphases and difficulties of detection task,and the present situation of development are put forward,with a time line for Convolution Neural Network(CNN)under the vehicle object detection algorithm is summarized,2 mainstream based on candidate box and the vehicle object detection algorithm based on regression are concluded,Finally,the development prospect of current object detection algorithm is prospected.
作者
苏山杰
陈俊豪
张之云
Su Shanjie;Chen Junhao;Zhang Zhiyun(School of Mechatronics and Vehicle Engineering,Chongqing Jiaotong University,Chongqing 400074)
出处
《汽车文摘》
2022年第8期14-23,共10页
Automotive Digest
关键词
深度学习
车辆目标检测
卷积神经网络
轻量化网络
Deep learning
Vehicle object detection
Convolutional Neural Network(CNN)
Lightweight network
