Cong Phuc Nguyen, V. Nguyen, Duc Dung Tran, A. Nguyen, N. P. Dao, D. Tran, Joo-Ho Lee, Anh Quang Nguyen
{"title":"Multi-task Deep-Learning Vehicle Detection and Tracking based on Aerial Views from UAV","authors":"Cong Phuc Nguyen, V. Nguyen, Duc Dung Tran, A. Nguyen, N. P. Dao, D. Tran, Joo-Ho Lee, Anh Quang Nguyen","doi":"10.1109/ATC55345.2022.9942962","DOIUrl":null,"url":null,"abstract":"For vehicle management systems, vehicle detection, tracking, and recognition which provides statistical information on the number of vehicles and their characters are essential task, however, not only the vehicles themselves but also its characteristic such as types, and colors,.. is important for management. Besides, not only fixed traffic cameras, autonomous UAVs with cameras also can bring more flexibility and extend the management areas from an aerial view. In this paper, we propose and implement a real-time multi-task deep-learning for four-wheel vehicle detection, classification, and tracking system on UAVs. A dataset for detection and multi-task classification including multiple colors and types is built. To archive real-time the operation, we used the ByteTrack algorithm with YOLOv5, which is more efficient for mobile and embedded vision applications. Experimental results achieved high accuracy, more than 90% in multi-task classifying, and 56.21 HOTA tracking evaluation metrics on our created test set. The processing time is fast enough for real-time operation, 14 FPS for the detector, and 64 FPS for classification based on an embedded computer, Jetson Nano.","PeriodicalId":135827,"journal":{"name":"2022 International Conference on Advanced Technologies for Communications (ATC)","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2022-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2022 International Conference on Advanced Technologies for Communications (ATC)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/ATC55345.2022.9942962","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
For vehicle management systems, vehicle detection, tracking, and recognition which provides statistical information on the number of vehicles and their characters are essential task, however, not only the vehicles themselves but also its characteristic such as types, and colors,.. is important for management. Besides, not only fixed traffic cameras, autonomous UAVs with cameras also can bring more flexibility and extend the management areas from an aerial view. In this paper, we propose and implement a real-time multi-task deep-learning for four-wheel vehicle detection, classification, and tracking system on UAVs. A dataset for detection and multi-task classification including multiple colors and types is built. To archive real-time the operation, we used the ByteTrack algorithm with YOLOv5, which is more efficient for mobile and embedded vision applications. Experimental results achieved high accuracy, more than 90% in multi-task classifying, and 56.21 HOTA tracking evaluation metrics on our created test set. The processing time is fast enough for real-time operation, 14 FPS for the detector, and 64 FPS for classification based on an embedded computer, Jetson Nano.