一种基于LSTM的视频车辆检测算法  

A LSTM-based method for vehicle detection in video

作　　者：李岁缠 陈锋[1] Li Suichan;Chen Feng(School of Information Science and Technology,University of Science and Technology of China,Hefei 230027,Chin)

机构地区：[1]中国科学技术大学信息科学技术学院,安徽合肥230027

出　　处：《信息技术与网络安全》2018年第7期54-57,共4页Information Technology and Network Security

摘　　要：针对视频车辆检测问题,提出了一种基于LSTM的视频车辆检测算法模型。该算法接受视频序列作为输入,先利用卷积网络提取视频帧的空间特征,然后利用LSTM模块得到时间维度的特征,最后利用全卷积网络预测最终的检测结果。将所提算法与其他典型的算法进行比较,实验结果表明所提算法具有更好的检测准确率,同时检测速度也更快。In this paper,a LSTM-based method was proposed for vehicle detection in video. This method takes multiple video frames as input and generates multiple spatial feature maps,and then spatial feature maps are fed to LSTM sub-model to capture temporal features. These temporal features are fed to full convolution networks to get final detection results. We compare the proposed method with other typical methods.The results show the proposed method yields better performance as well as higher processing speed.

关 键 词：视频车辆检测 LSTM 时间维度特征 CNN 深度学习 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]