基于Cascade-Rcnn的行人检测  

Pedestrian detection method based on Cascade-Rcnn

作　　者：刘博文 彭祝亮 范程岸 Liu Bowen;Peng Zhuliang;Fan Cheng’an(School of Automation,Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]广东工业大学自动化学院

出　　处：《无线互联科技》2020年第2期15-17,共3页Wireless Internet Technology

基　　金：广东省科技计划项目；项目编号：2019B010147001;广州市科技计划项目；项目编号：201902020014

摘　　要：Faster Rcnn是目标检测领域中精确度较高、使用范围较广的一个经典算法,而Cascade Rcnn是借鉴Faster Rcnn改进的。文章将Cascade Rcnn的方法应用于行人检测中,利用深层卷积神经网络提取图像特征,通过RPN提取可能含有行人的区域,利用多层级联的检测器对目标区域进行判别和分类,在数据集中进行了检测验证。实验结果表明,相比基于Faster-rcnn的行人检测方法,在测试集上检测准确度达到了66.2%,其检测效果更好。Faster Rcnn is a classic algorithm with high accuracy and wide range of target detection, Cascade Rcnn is improved by using Faster Rcnn. This paper applies Cascade Rcnn’s method to pedestrian detection, uses deep convolutional neural network to extract image features, and extract areas that may contain pedestrians by RPN. The region uses a multi-level cascaded detector to discriminate and classify the target area, and performs verification in the data set. The experimental results show that compared with the Faster-rcnn-based pedestrian detection method, the detection accuracy on the test set reaches 66.2%, and the detection effect is better.

关 键 词：行人检测 深度学习 目标检测 深层卷积神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]