第 44 卷 第 12 期
2018 年 12 月
自
动
化
学
报
Vol. 44, No. 12
ACTA AUTOMATICA SINICA
December, 2018
改进的 YOLO 特征提取算法及其在服务机器人
隐私情境检测中的应用
苏志东 1
杨观赐 1
杨 静 1
陈占杰 1
摘
要
为了提高 YOLO 识别较小目标的能力, 解决其在特征提取过程中的信息丢失问题, 提出改进的 YOLO 特征提取算
法. 将目标检测方法 DPM 与 R-FCN 融入到 YOLO 中, 设计一种改进的神经网络结构, 包含一个全连接层以及先池化再卷
积的特征提取模式以减少特征信息的丢失. 然后, 设计基于 RPN 的滑动窗口合并算法, 进而形成基于改进 YOLO 的特征提取
算法. 搭建服务机器人情境检测平台, 给出服务机器人情境检测的总体工作流程. 设计家居环境下的六类情境, 建立训练数据
集、验证数据集和 4 类测试数据集. 测试分析训练步骤与预测概率估计值、学习率与识别准确性之间的关系, 找出了适合所
提出算法的训练步骤与学习率的经验值. 测试结果表明: 所提出的算法隐私情境检测准确率为 94.48 %, 有较强的识别鲁棒性.
最后, 与 YOLO 算法的比较结果表明, 本文算法在识别准确率方面优于 YOLO 算法.
关键词 YOLO, 特征提取算法, 服务机器人, 隐私情境检测, 智能家居
引用格式 杨观赐, 杨静, 苏志东, 陈占杰. 改进的 YOLO 特征提取算法及其在服务机器人隐私情境检测中的应用. 自动化学
报, 2018, 44(12): 2238−2249
DOI 10.16383/j.aas.2018.c170265
An Improved YOLO Feature Extraction Algorithm and Its Application to
Privacy Situation Detection of Social Robots
YANG Guan-Ci1
YANG Jing1
SU Zhi-Dong1
CHEN Zhan-Jie1
Abstract To address the limitation of YOLO algorithm in recognizing small objects and information loss during feature
extraction, we propose FYOLO, an improved feature extraction algorithm based on YOLO. The algorithm uses a novel
neural network structure inspired by the deformable parts model (DPM) and region-based fully convolutional networks
(R-FCN). A sliding window merging algorithm based on region proposal networks (RPN) is then combined with the
neural network to form the FYOLO algorithm. To evaluate the performance of the proposed algorithm, we develop a
social robot platform for privacy situation detection. We consider six types of situations in a smart home and prepare
three datasets including training dataset, validation dataset, and test dataset. Experimental parameters such as training
step and learning rate are set in terms of their relationships with the prediction accuracy. Extensive privacy situation
detection experiments on the social robot show that FYOLO is capable of recognizing privacy situations with an accuracy
of 94.48 %, indicating the good robustness of our FYOLO algorithm. Finally, the comparison results between FYOLO
and YOLO show that the proposed FYOLO outperforms YOLO in recognition accuracy.
Key words YOLO, feature extraction algorithm, social robot, detection of privacy situations, smart homes
Citation Yang Guan-Ci, Yang Jing, Su Zhi-Dong, Chen Zhan-Jie. An improved YOLO feature extraction algorithm
and its application to privacy situation detection of social robots. Acta Automatica Sinica, 2018, 44(12): 2238−2249
越来越多的智能家居系统和服务机器人广泛使
收稿日期 2017-05-15 录用日期 2017-08-29
Manuscript received May 15, 2017; accepted August 29, 2017
国家自然科学基金 (61863005, 61640209), 贵州省科技计划项目 (黔
用摄像头 这会引入隐私泄漏风险 是阻碍此类系统
推广的最大障碍之一[1] 前期问卷调查表明[2] 对隐
科合人字 (2015) 13, 黔科合 JZ 字 [2014] 2004, 黔科合 LH 字 [2016]
私内容有符合人心理需求反应的系统 可改善用户
7433, 黔科合平台人才 [2018] 5702), 贵州省教育厅研究生教改重点课
体验感受 如何识别与保护具有视觉设备的服务机
等[3]
设
题 (黔教研合 JG 字 [2015] 002) 资助
Supported by National Natural Science Foundation of China
器人的隐私数据是值得研究的问题
(61863005, 61640209), Science and Technology Foundation of
Guizhou Province ((2015) 13, JZ [2014] 2004, LH [2016] 7433,
PTRC [2018] 5702), and Graduate Education Reform Fund of
计了一种智能家居环境中隐私与安全框架
等[4] 通过分析智能家居环境下各系统间的安全和隐
Education Bureau of Guizhou Province (JG [2015] 002)
本文责任编委 胡清华
私与互信风险 研究了高度依赖法律支持的隐私控
Recommended by Associate Editor HU Qing-Hua
1. 贵州大学现代制造技术教育部重点实验室 贵阳 550025
1. Key Laboratory of Advanced Manufacturing Technology of
Ministry of Education, Guizhou University, Guiyang 550025
全部评论(0)