Detecting Internet-Scale Surveillance Devices Using RTSP Recessive Features

Zhaoteng Yan, Zhi Li, Wenping Bai, Nan Yu, Hongsong Zhu, Limin Sun

Abstract: In recent years, fingerprinting online surveillance devices has been a hot research topic. However, large-scale devices still can not be identified their brands in previous studies and mainstream search engines. In this work, we propose a novel neural network-based approach for automatically discovering surveillance devices and identifying their brands in cyberspace. Moreover, by using the deep semi-supervised learning algorithm, the most unlabeled samples with new-explored recessive features can be learned of RTSP protocol. In the global IPv4 space, we implement an evaluation on 3, 123, 489 active RTSP-hosts for training and testing. The experimental results demonstrate our approach can discover 2, 803, 406 surveillance devices, which are eight times and three times more than those discovered by Shodan and Zoomeye. Moreover, the number of identified brand-level devices by our approach is 2, 457, 661 devices with their brands, which is at least four times more than existing methods. The performance of these results with precision and recall can both achieve $$93\%$$ .