Chunhui Duan

Chunhui Duan

Auto-generated profile, based on publicly available information

Chunhui Duan

Beijing Institute of Technology, Beijing, China

Research Focus & Metrics

Academic Career Length

9Years (2018-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
26 February 2024 in IEEE Transactions on Circuits and Systems for Video Technology
Verified
Journal Article
Closed access
QFQihua Feng
Qihua Feng
PLPeiya Li
Peiya Li
ZLZhixun Lu
Zhixun Lu
CLChaozhuo Li
Chaozhuo Li
ZWZefang Wang
Zefang Wang
PYPhilip S. Yu
Philip S. Yu
Image retrieval systems help users to browse and search among extensive images in real time. With the rise of cloud computing, retrieval tasks are usually outsourced to cloud servers. However, the cloud scenario brings a daunting challenge of privacy protection as cloud servers cannot be fully trusted. To this end, image-encryption-based privacy-preserving image retrieval (PPIR) schemes have been developed, which first extract features from cipher-images, and then build retrieval models based on these features. Yet, most existing PPIR approaches extract shallow features and design trivial unsupervised retrieval models, resulting in insufficient expressiveness for the cipher-images. In this paper, we propose a novel paradigm named Encrypted Vision Transformer (EViT), which advances the discriminative representations capability of cipher-images. First, to capture comprehensive ruled information, we extract multi-level local length sequence and global Huffman-Code frequency features from the cipher-images which are encrypted by permutation encryption, sign encryption, and stream cipher during the JPEG compression process. Second, we design the modified self-supervised Vision Transformer with Huffman-embedding and propose two robust data augmentations on cipher-images to improve representation power of the retrieval model. Moreover, our proposal can be easily adapted to unsupervised or supervised settings. Extensive experiments reveal that EViT achieves both excellent encryption and retrieval performance, outperforming current schemes in terms of retrieval accuracy by large margins while protecting image privacy effectively. Code is publicly available at https://github.com/onlinehuazai/EViT.

25 October 2018 in IEEE Transactions on Mobile Computing
Verified
Journal Article
Closed access
CDChunhui Duan
Chunhui Duan
LYLei Yang
Lei Yang
QLQiongzheng Lin
Qiongzheng Lin
YLYunhao Liu
Yunhao Liu
LXLei Xie
Lei Xie
Conventional spinning inspection systems, equipped with separated sensors and communication modules, are either very expensive and/or suffering from occlusion and narrow field of view. The recently proposed RFID-based sensing solution draws much attention due to its intriguing features, such as being cost-effective, applicable to occluded objects and auto-identification, etc. However, this solution only works in quiet settings where both the reader and spinning object remain absolutely stationary, as their shaking would ruin the periodicity and sparsity of the spinning signal, making it impossible to be recovered. To overcome such limitation, this work introduces Tagtwins, a robust spinning sensing system that can work in noisy settings. It addresses the challenge by attaching dual RFID tags on the spinning surface and developing a new formulation of spinning signal that is shaking-resilient. Our main contribution lies in two newly developed techniques. First, we propose relative spinning signal using dual tags' readings and analytically demonstrate its feasibility in various settings. Second, we introduce dual compressive reading to inspect high-frequency spinning with relatively low reading rate of RFIDs. We have implemented Tagtwins with commercial RFID devices and evaluated it extensively. Experimental results show that Tagtwins can inspect the rotation frequency with high accuracy and robustness.

23 January 2018 in IEEE Transactions on Mobile Computing
Verified
Journal Article
Closed access
CDChunhui Duan
Chunhui Duan
LYLei Yang
Lei Yang
QLQiongzheng Lin
Qiongzheng Lin
YLYunhao Liu
Yunhao Liu
Recent years have witnessed the advance of RFID-based localization techniques that demonstrate high precision. Many efforts have been made locating RFID tags accurately with a mandatory assumption that the RFID reader's position is known in advance. Unfortunately, calibrating reader's location manually is always time-consuming and laborious in practice. In this paper, we present Tagspin, an approach using COTS tags to pinpoint the reader (antenna) quickly and easily with high accuracy. Tagspin enables each tag to emulate a circular antenna array by uniformly spinning on the edge of a rotating disk. We design an SAR-based method for estimating the angle spectrum of the target reader. Compared to previous AoA-based techniques, we employ an enhanced power profile modeling the relative signal power received from the reader along different spatial directions, which is more accurate and immune to ambient noise as well as measurement errors caused by hardware characteristics. Besides, we find that tag's phase measurements in practice are related to its orientation. To the best of our knowledge, we are the first to point out this fact and quantify the relationship between them. By calibrating the phase shifts caused by orientation, the positioning accuracy can be improved by 3:7×. We have implemented Tagspin with COTS RFID devices and evaluated it extensively. Experimental results show that Tagspin achieves mean accuracy of 7:3 cm with standard deviation of 1:8 cm in 3D space.

14 February 2019 in IEEE/ACM Transactions on Networking
Verified
Journal Article
Closed access
QLQiongzheng Lin
Qiongzheng Lin
LYLei Yang
Lei Yang
CDChunhui Duan
Chunhui Duan
ZAZhenlin An
Zhenlin An
Deployment of billions of commercial off-the-shelf (COTS) radio frequency identification (RFID) tags has drawn much of the attention of the research community because of the performance gaps of current systems. In particular, hash-enabled protocol (HEP) is one of the most thoroughly studied topics in the past decade. HEPs are designed for a wide spectrum of notable applications (e.g., missing detection) without need to collect all tags. HEPs assume that each tag contains a hash function, such that a tag can select a random but predictable time slot to reply with a one-bit presence signal that shows its existence. However, the hash function has never been implemented in COTS tags in reality, which makes HEPs a ten-year untouchable mirage. This paper designs and implements a group of analog on-tag hash primitives (called Tash) for COTS Gen2-compatible RFID systems, which moves prior HEPs forward from theory to practice. In particular, we design three types of hash primitives, namely, tash function, tash table function, and tash operator. All of these hash primitives are implemented through the selective reading, which is a fundamental and mandatory functionality specified in Gen2 protocol, without any hardware modification and fabrication-a feature allowing zero-cost fast deployment on billions of Gen2 tags. We further apply our hash primitives in one typical HEP application (i.e., missing detection) to show the feasibility and effectiveness of Tash. Results from our prototype, which is composed of one ImpinJ reader and 3000 Alien tags, demonstrate that the new design lowers 70% of the communication overhead in the air. The tash operator can additionally introduce an overhead drop of 29.7%.

13 August 2018 in IEEE Transactions on Mobile Computing
Verified
Journal Article
Closed access
QLQiongzheng Lin
Qiongzheng Lin
LYLei Yang
Lei Yang
CDChunhui Duan
Chunhui Duan
YLYunhao Liu
Yunhao Liu
RFID systems are currently supplemented with various interesting sensing functions. All these sensing applications forcedly require much higher reading rate. However, COTS RFID systems suffer from an extremely low individual reading rate when multiple tags are present, due to their intense channel contention in the link layer. In this work, we present a holistic system, called Tagwatch, a rate-adaptive reading system for COTS RFID devices. This work revisits the reading rate from a distinctive perspective: mobility. We observe that the reading demands of mobile tags are considerably more urgent than those of stationary tags because the states of the latter nearly remain unchanged; meanwhile, only a few tags (e.g., < 20%) are actually in motion despite the existence of a massive amount of tags in practice. Thus, Tagwatch adaptively improves the reading rates for mobile tags by cutting down the readings of stationary tags. Our main contribution is a two-phase reading design, wherein the mobile tags are discriminated in the first phase and exclusively read in the second phase. Results from our microbenchmark analysis demonstrate that the new design outperforms the reading rate by 3.2% when 5% of tags are moving.

27 December 2022 in IEEE Internet of Things Journal
Verified
Journal Article
Closed access
YWYusen Wang
Yusen Wang
FLFan Li
Fan Li
YXYadong Xie
Yadong Xie
CDChunhui Duan
Chunhui Duan
YWYu Wang
Yu Wang
Sign language is expressed by movements of the hands and facial expressions, which is mainly used by the deaf community. Although some gesture recognition methods are put forward, they possess different defects and are not applicable to deal with the sign language recognition (SLR) problem. In this article, we propose an end-to-end American SLR system with built-in speakers and microphones in smartphones, which enables SLR at both word level and sentence level. The high-level idea is to use the inaudible acoustic signal to estimate channel information and capture the sign language in real time. We use channel impulse response to represent each sign language gesture, which can realize finger-level recognition. We also pay attention to conversion movements between two words and treat them as an additional label when training the sentence-level classification model. We implement a prototype system and run a series of experiments that demonstrate the promising performance of our system. Experimental results show that our approach can achieve an accuracy of 97.2% at word-level recognition and word error rate of 0.9% at sentence-level recognition, respectively.