Fan Wu, Yantao Zong, Rui Zhao, Tzuyang Yu, Xiaqing Tang, Ximing He
{"title":"Visual odometery for UAV navigation based on deep learning","authors":"Fan Wu, Yantao Zong, Rui Zhao, Tzuyang Yu, Xiaqing Tang, Ximing He","doi":"10.1145/3501409.3501673","DOIUrl":null,"url":null,"abstract":"Taking UAV as the application background, this paper studies the visual odometery for deep learning in UAV navigation. Firstly, the dataset FLYING is made through UAV aerial photography. Secondly, pretraining and testing the established G-LSTM VO and attention VO models through the KITTI dataset. Thirdly, by means of transfer learning, training and testing the pre trained model based on FLYING dataset. Finally, the model is tested. The performance of the model in UAV mission is analyzed from the aspects of trajectory, pose estimation accuracy and algorithm time-consuming. The experimental results show that the monocular visual odometery method based on depth neural network is effective in the field of UAV navigation.","PeriodicalId":191106,"journal":{"name":"Proceedings of the 2021 5th International Conference on Electronic Information Technology and Computer Engineering","volume":"5 2","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2021-10-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Proceedings of the 2021 5th International Conference on Electronic Information Technology and Computer Engineering","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1145/3501409.3501673","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
Taking UAV as the application background, this paper studies the visual odometery for deep learning in UAV navigation. Firstly, the dataset FLYING is made through UAV aerial photography. Secondly, pretraining and testing the established G-LSTM VO and attention VO models through the KITTI dataset. Thirdly, by means of transfer learning, training and testing the pre trained model based on FLYING dataset. Finally, the model is tested. The performance of the model in UAV mission is analyzed from the aspects of trajectory, pose estimation accuracy and algorithm time-consuming. The experimental results show that the monocular visual odometery method based on depth neural network is effective in the field of UAV navigation.