| 摘要: |
| 对于行人的再识别研究大多采用图像处理和计算机视觉领域的相关方法，在社会治安领域和商业领域内受到了越来越多的关注.从信息检索的角度出发，提出了一种端到端的深度学习框架，对匿名化的基于位置的服务（LBS）数据进行用户再识别.首先，该框架采用嵌入网络对输入的位置序列及其对应的时间序列进行编码；然后采用递归循环网络对用户每天的历史轨迹进行编码；随后连接注意力机制网络，对需要比较的两条轨迹进行重要权重计算；最后得出其相似度.实验结果表明：相较于计算轨迹之间向量距离的传统方法，此模型考虑了用户的时空位置信息，可以更加准确地计算轨迹序列之间的相似度，在某城市匿名化的LBS数据集上，对不同数量的用户重识别准确率较高. |
| 关键词: 轨迹重识别 注意力机制网络 深度学习 |
| DOI：10.3969/J.ISSN.1000-5137.2021.01.016 |
| 分类号:TP399 |
| 基金项目: |
|
| An end-to-end deep learning method for individual travel path re-identification |
|
LU Jiashuang1, WANG Bin1, ZHAI Xi2
|
|
1.College of Information Mechanical and Electrical Engineering, Shanghai Normal University, Shanghai 200234, China;2.Shanghai Traffic Information Center, Shanghai Urban and Rural Construction and Traffic Development Research Institute, Shanghai 200003, China
|
| Abstract: |
| A large number of researches on pedestrian re-identification based on the methods of image processing and computer vision were getting more and more attention in the field of social security and business. From the perspective of information retrieval,an end-to-end deep learning framework was proposed for user re-identification of anonymous location based services (LBS)data in this paper. Firstly, the embedded network was used to encode the input spatial sequence and the corresponding temporal sequence. Secondly,the recurrent network was adopted to encode the user's daily history trajectory. Thirdly,the attention mechanism network was connected to calculate the importance weight of the two trajectories to be compared, and finally the similarity of the two trajectories was obtained. The experimental results showed that this model was able to take the user's spatial-temporal position information into account, and achieve more accurate similarity between trajectory sequences compared with the traditional method of calculating the vector distance between trajectories. The re-identification accuracy of different number of users on the anonymous LBS dataset of a city was significantly improved. |
| Key words: trajectory re-identification attention-based network deep learning |
Editorial Board