| 基于深度学习的监控视频车辆实时监测 |
| |
| 引用本文： | 张文辉.基于深度学习的监控视频车辆实时监测[J].教育技术导刊,2019,18(7):37-40. |
| |
| 作者姓名： | 张文辉 |
| |
| 作者单位： | 广东工业大学 自动化学院，广东 广州 510006 |
| |
| 摘 要： | 针对传统车辆检测算法不能自适应地完成复杂道路场景变化下提取车辆特征的问题，结合焦点损失、K-means聚类与mobilenet网络，提出改进的RFB-VGG16与RFB-MobileNet模型进行车辆检测。从开源数据集UA-DETRAC的24个视频中每隔一定帧数抽取8 209张已标注的图片构成数据集，在相同的超参数与训练策略下，改进后RFB-VGG16网络的AP值比原模型提高了3.2%。基于mobilenet网络重新设计RFB骨架网络，使RFB-MobileNet模型在牺牲一定性能的情况下，具有更快的检测速度，能较好地满足监控视频对车辆检测实时性的要求。
|
| 关 键 词： | 深度学习 车辆检测 焦点损失 RFBNet K-means |
| 收稿时间： | 2018-12-06 |
Real-time Vehicle Detection in Surveillance Video Based on Deep Learning |
| |
| ZHANG Wen-hui.Real-time Vehicle Detection in Surveillance Video Based on Deep Learning[J].Introduction of Educational Technology,2019,18(7):37-40. |
| |
| Authors: | ZHANG Wen-hui |
| |
| Institution: | School of Automation，Guangdong University of Technology，Guangzhou 510006，China |
| |
| Abstract: | Aiming at the problem that the traditional vehicle detection algorithm can not adaptively extract the vehicle characteristics under the complex road scene change， this paper combines the focus loss， K-means clustering and mobilenet network， and proposes the improved RFB-VGG16 and RFB-MobileNet models for the vehicle detection. First， 8209 images of the labeled images are extracted from the 24 videos of the open source dataset UA-DETRAC. Second，Under the same hyperparameters and training strategies， the AP value of the improved RFB-VGG16 network is 3.2% higher than the original model. Redesigning the RFB's skeleton network based on the mobilenet network enables the RFB-MobileNet model to have a faster detection speed at the expense of a little performance， thus meeting the real-time requirements of vehicle detection in surveillance video. |
| |
| Keywords: | deep learning vehicle detection focus loss RFBNet K-means |
|
| 点击此处可从《教育技术导刊》浏览原始摘要信息 |
| 点击此处可从《教育技术导刊》下载免费的PDF全文 |
