摘要
针对传统行人检测器鲁棒性差,定位精度差且漏检率较高的问题,提出一种基于YOLOV3网络结构行人检测方法。结合行人尺寸特点和改变卷积层的数量,聚类选取恰当的候选框,改进YOLOV3网络结构,得到适用于行人检测的网络结构。实验结果表明,与HOG+SVM、Faster R-CNN、YOLO等主流方法比较,改进的YOLOV3行人检测方法对于定位的准确性和精确度有一定的提升。
Due to the poor robustness of traditional pedestrian detector,poor positioning accuracy,and high missed detection rate,the paper propose a pedestrian detection method based on YOLOV3 network structure.We attain a network structure suitable for pedestrian detection by combining features of pedestrian size,changing the number of convolutional layers,clustering selects appropriate candidate frames,and improving YOLOV3 network structure.The results suggest that compared with the mainstream methods such as HOG+SVM,Faster R-CNN and YOLO,the method proposed by this paper achieve improvement on the accuracy and accuracy of positioning to a certain extent.
作者
孟本成
MENG Bencheng(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China)
出处
《电视技术》
2019年第9期6-9,46,共5页
Video Engineering
关键词
行人数据集
行人检测
深度学习
聚类
YOLO
pedestrian dataset
pedestrian detection
deep learning
clustering
YOLO
