摘要
前方车辆识别是实现自动驾驶环境感知中的最重要的课题之一,目标检测需要高的检测精度和定位精度以及实时性和鲁棒性。目标检测的传统算法中,典型代表有Haar特征+Adaboost算法,Hog特征+Svm算法,Dpm算法。深度学习的目标检测典型代表有RCNN系列,YOLO系列,SSD,YOLO是目前最快的目标检测的卷积神经网络算法。通过YOLO算法对公开数据集中车辆目标进行测试,对不同环境中的采集图像进行测试,实验结果表明YOLO算法能够满足车辆检测的实时性和精度的要求,说明该方法可行。
In front of the vehicle identification is to realize the automatic driving one of the most important subject in environ⁃mental awareness,target detection need high detection accuracy and positioning accuracy and real-time performance and robust⁃ness of target detection in the traditional algorithm.The typical representatives are Haar feature+Adaboost algorithm,Hog feature+Svm algorithm and Dpm algorithm.Typical representatives of deep learning target detection are RCNN series,series of YOLO,SSD.YOLO is currently the fastest convolution neural network algorithm of target detection.Through YOLO algorithm,vehicle targets in the open data set are tested,and images collected in different environments are tested.The experimental results show that the YOLO algorithm can meet the requirements of real-time and accuracy of vehicle detection,indicating that the method is feasible.
作者
何旭光
罗一平
江磊
HE Xuguang;LUO Yiping;JIANG Lei(School of Mechanical and Automotive Engineering,Shanghai University of Engineering and Technology,Shanghai 201620)
出处
《舰船电子工程》
2021年第1期137-139,共3页
Ship Electronic Engineering
关键词
车辆识别
卷积神经网络
YOLO
vehicle identification
convolutional neural network
YOLO
