摘要
针对实际交通场景下道路目标检测时存在检测精度低、检测速度慢以及难以检测小目标的问题,构建一种基于Faster R-CNN的快速、精确道路目标检测算法。该算法包括一个精确目标区域网络(AORN)和一个目标属性学习网络(OALN)。通过引入反卷积结构,设计AORN网络和OALN网络的损失函数,提高小目标的检测性能,为加快算法的计算速度,AORN和OALN交替优化、联合训练。实验结果表明,其测试的平均准确率较先进的目标检测算法Faster R-CNN 提高了0.15,检测速度提高了3 fps。
Aiming at the problems of road object detection including low detection accuracy,low detection speed and difficult detection of small object,an algorithm based on Faster region-based convolutional neural network (r-CNN) was proposed. The algorithm included two parts,namely the accurate object proposal network (AORN) and the object attribute learning network (OALN). By introducing a de-convolution structure and designing the loss functions of both the AORN and the OALN,the detection performance of small object was improved. To speed up the calculation of the algorithm,the AORN and OALN were alternately optimized and jointly trained. The experimental results show that the average precision of the test results is improved by 0.15 and the detection speed is increased by 3 fps higher than that of the state-of-the-art object detection algorithm Faster R-CNN ,respectively.
作者
张庆辉
万晨霞
秦淑英
卞山峰
ZHANG Qing-hui;WAN Chen-xia;QIN Shu-ying;BIAN Shan-feng(College of Information Science and Engineering,Henan University of Technology,Zhengzhou 450001,China;Hebi Second TV Relay Station,Hebi 458000,China)
出处
《计算机工程与设计》
北大核心
2019年第7期2052-2058,共7页
Computer Engineering and Design
基金
国家自然科学基金项目(U1404617)
河南省科技创新人才计划杰出青年基金项目(174100510011)
河南省高校科技创新团队基金项目(16IRTSTHN026)
关键词
深度学习
卷积神经网络
道路图像
目标检测
目标区域网络
deep learning
convolutional neural network
road image
object detection
object proposal network
