Article
Computer Science, Information Systems
Muhammad Ali Farooq, Peter Corcoran, Cosmin Rotariu, Waseem Shariff
Summary: This research evaluates and modifies AI-based smart thermal perception systems for ADAS, providing reliable thermal sensing input by training networks on public datasets and testing with various approaches. In addition, a new model ensemble-based inference engine is proposed and tested on novel test data, while optimizations are made to reduce inference time and assess real-time onboard installations.
Article
Computer Science, Information Systems
Diego Renan Bruno, Rafael A. Berri, Felipe M. Barbosa, Fernando S. S. Osorio
Summary: Autonomous mobile robots rely on complex computational techniques and intelligent sensing to navigate and operate safely in dynamic environments. Computer vision systems play a crucial role in detecting and tracking obstacles, improving traffic safety with intelligent robotic vehicles.
Article
Chemistry, Analytical
Hamza Nadeem, Kashif Javed, Zain Nadeem, Muhammad Jawad Khan, Saddaf Rubab, Dong Keon Yon, Rizwan Ali Naqvi
Summary: Hundreds of people are injured or killed in road accidents due to factors like driver inattentiveness. This study proposes a computer vision-based solution using deep learning models to detect and recognize road features such as traffic types and signs. The models achieved state-of-the-art results, providing a benchmark for improving traffic situations and enabling future technological advances.
Article
Chemistry, Multidisciplinary
Chathura Neelam Jaikishore, Gautam Podaturpet Arunkumar, Ajitesh Jagannathan Srinath, Harikrishnan Vamsi, Kirtaan Srinivasan, Rishabh Karthik Ramesh, Kathirvelan Jayaraman, Prakash Ramachandran
Summary: This study proposes an ADAS device based on the YOLO model for detecting and marking road obstacles. The YOLOv3 model performs exceptionally well in the performance evaluation.
APPLIED SCIENCES-BASEL
(2022)
Article
Engineering, Civil
Muhammad Mobaidul Islam, Abdullah Al Redwan Newaz, Ali Karimoddini
Summary: This paper proposes a novel fusion framework that combines asymmetric inferences from object detectors and semantic segmentation networks for jointly detecting multiple pedestrians. By introducing a consensus-based scoring method to fuse pair-wise pixel-relevant information, the final confidence scores are boosted with low runtime overhead through parallel implementation.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Lukas Ewecker, Ebubekir Asan, Lars Ohnemus, Sascha Saralajew
Summary: In recent years, computer vision algorithms have greatly improved, enabling the rapid development of technologies like autonomous driving. However, current algorithms have a limitation in that they rely on directly visible objects. This paper presents a complete system that can detect light artifacts caused by the headlights of oncoming vehicles to proactively detect approaching vehicles. By deploying the algorithm in a test vehicle, it can control the glare-free high beam system. This research aims to bridge the performance gap between human behavior and computer vision algorithms for autonomous driving.
Article
Computer Science, Artificial Intelligence
Jian-Xun Mi, Xu-Dong Wang, Li -Fang Zhou, Kun Cheng
Summary: Deep learning plays a critical role in artificial intelligence applications, particularly in the processing of image or video data for efficient execution. However, the complex structure of deep networks makes them susceptible to attacks, posing a risk to object detection tasks, which are extensively integrated into our lives and can result in significant loss. Adversarial example attacks have emerged as an effective and understandable approach to generate perturbations. This survey reviews existing adversarial example attacks in object detection, discussing their similarities, differences, and potential directions for adversarial defenses in future studies.
Article
Chemistry, Analytical
Xuelin Zhang, Donghao Zhang, Alexander Leye, Adrian Scott, Luke Visser, Zongyuan Ge, Paul Bonnington
Summary: This paper focuses on improving the performance of scientific instrumentation that uses glass spray chambers for sample introduction, by detecting incidents using deep convolutional models. The indicators of poor quality sample introduction include the formation of liquid beads and flooding in the spray chamber. The proposed frameworks for detecting these incidents leverage modern deep learning architectures and expert knowledge, achieving high accuracy and real-time implementation.
Article
Computer Science, Information Systems
Ying-Cheng Lin, Ping-Yen Chiang, Shaou-Gang Miaou
Summary: This study explores the feasibility of using heterogeneous image fusion to improve the object detection performance of advanced driver assistance systems (ADAS). The fusion of infrared and visible images can improve the object detection performance of deep learning networks. Therefore, an image-fusion approach that combines alignment and fusion methods is proposed as an effective solution for ADAS applications.
Article
Computer Science, Information Systems
Yingxin Qin, Kejia Zhang, Haiwei Pan
Summary: Deep learning security has gained significant attention in recent years, particularly in object detection where attackers use adversarial examples to deceive models. However, the current research on adversarial patches for object detectors has limitations regarding occlusion. To address this, a two-stage method called TS-GAN, based on Generative Adversarial Network, is proposed. It trains the generator using occlusion rules to simulate different occlusion scenarios, leading to the generation of effective adversarial patches.
COMPUTERS & SECURITY
(2023)
Article
Computer Science, Information Systems
Yaoyuan Zhang, Yu-an Tan, Mingfeng Lu, Lu Liu, Dianxin Wang, Quanxing Zhang, Yuanzhang Li
Summary: Recent works have shown that deep learning models are vulnerable to adversarial examples, limiting their application in security-critical systems. This paper proposes adversarial distillation to interpret the vulnerability and demonstrates the effectiveness of using adversarial features to improve model generalization on adversarial datasets. Experimental evaluations show that the adversarial distillation model has excellent generalization performance compared to normally trained models.
JOURNAL OF INFORMATION SECURITY AND APPLICATIONS
(2023)
Article
Engineering, Electrical & Electronic
Da-Wei Jaw, Shih-Chia Huang, I-Chuan Lin, Cheng Zhang, Ching-Chun Huang, Sy-Yen Kuo
Summary: Advanced object detection techniques have been widely studied and successfully applied in real-world applications. However, they face challenges in nighttime image detection, especially in low-luminance conditions. In this study, a lightweight framework using generative adversarial networks (GANs) is proposed for multidomain object detection, which includes feature domain transformation and a training policy to achieve luminance-invariant feature extraction. The proposed method outperforms existing algorithms with a 9.95% improvement in average precision, without incurring additional computational costs.
IEEE SENSORS JOURNAL
(2023)
Article
Engineering, Electrical & Electronic
Qinghai Lang, Lei Zhang, Wenxu Shi, Weijie Chen, Shiliang Pu
Summary: The Implicit Domain-invariant Faster R-CNN (IDF) is proposed to address the problem of implicit domain-invariant features caused by the multimodal structure of target distribution. By using a non-adversarial domain discriminator, dual attention mechanism, and selective feature perception, IDF outperforms other state-of-the-art domain adaptive object detectors on benchmark datasets.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Computer Science, Information Systems
Ke He, Dan Dongseong Kim, Muhammad Rizwan Asghar
Summary: Network-based Intrusion Detection System (NIDS) is vital for defending against network attacks, but it is susceptible to adversarial attacks that manipulate input examples. This article reviews the literature on NIDS, adversarial attacks, and defence mechanisms, highlighting the challenges in launching and detecting adversarial attacks against NIDS.
IEEE COMMUNICATIONS SURVEYS AND TUTORIALS
(2023)
Article
Computer Science, Artificial Intelligence
Qianxi Zhao, Liu Yang, Nengchao Lyu
Summary: Excessive stress leads to degraded driving performance and increases the risk of road accidents. This paper proposes a real-time driver stress detection method using deep learning and generative adversarial networks, focusing on analyzing pupillary response data. By establishing different models and utilizing data augmentation, the accuracy of stress detection and the recognition rate of minority categories are improved.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Article
Engineering, Biomedical
Junshi Liu, Xingda Qu, Michael H. Stone
SPORTS BIOMECHANICS
(2023)
Article
Psychology, Developmental
Zhong Zhao, Haiming Tang, Xiaobin Zhang, Zhipeng Zhu, Jiayi Xing, Wenzhou Li, Da Tao, Xingda Qu, Jianping Lu
Summary: This study used an eye tracker to record the gaze behavior of Chinese children with ASD and children with typical development during a conversation. The results showed that children with ASD paid less attention to the interlocutor's mouth and whole-face and more to the background. Moreover, gaze behavior varied with the conversational topic.
JOURNAL OF AUTISM AND DEVELOPMENTAL DISORDERS
(2023)
Review
Engineering, Electrical & Electronic
Caizhi Zhang, Weifeng Huang, Tong Niu, Zhitao Liu, Guofa Li, Dongpu Cao
Summary: Clustering is an unsupervised learning technology that groups information based on similarity measures. This paper introduces the concept of clustering and analyzes clustering technologies from traditional and modern perspectives. It summarizes the principles, advantages, and disadvantages of various traditional and modern clustering algorithms, and presents core elements such as similarity measures and evaluation index. The paper also lists specific applications of clustering algorithms in vehicles and highlights the future development of clustering in the era of big data.
AUTOMOTIVE INNOVATION
(2023)
Article
Engineering, Industrial
Tingru Zhang, Xing Liu, Weisheng Zeng, Da Tao, Guofa Li, Xingda Qu
Summary: This study compared the effects of three novel input modalities (touchscreen, speech-based, and gesture-based) on driving performance and driver visual behaviors. The results showed that touchscreen interaction had a negative impact on driving performance, while gesture-based interaction had a smaller but still significant crash risk. Speech-based interaction had the least influence on driving and visual performance. The effects of different modalities were robust across different non-driving related tasks (NDRTs).
APPLIED ERGONOMICS
(2023)
Article
Engineering, Electrical & Electronic
Guofa Li, Long Zhang, Ying Zou, Delin Ouyang, Yufei Yuan, Qiuyan Lian, Wenbo Chu, Gang Guo
Summary: This article examines the potential of using EEG signals from only one frequency band or from only a small subset of related electrode channels in recognizing driver vigilance state. The experimental results show that the recognition accuracy is higher when using EEG signals from the selected frequency band (i.e., alpha band) or the selected electrodes (i.e., T7, TP7, and CP1) than when using all the data. These results indicate that higher driver vigilance recognition accuracy can be achieved with much less amount of data, which would facilitate the development of wearable equipment based on EEG signals.
IEEE SENSORS JOURNAL
(2023)
Review
Engineering, Electrical & Electronic
Ashish Pandharipande, Chih-Hong Cheng, Justin Dauwels, Sevgi Z. Gurbuz, Javier Ibanez-Guzman, Guofa Li, Andrea Piazzoni, Pu Wang, Avik Santra
Summary: Automotive perception is crucial for achieving high levels of safety and autonomy in driving, involving the understanding of the external driving environment and the internal state of the vehicle cabin and occupants using sensor data. This article provides an overview of different sensor modalities commonly used for perception, along with data processing techniques. Critical aspects such as architectures, algorithms, and safety are discussed, with a focus on machine learning approaches. Future research opportunities in automotive perception are also outlined.
IEEE SENSORS JOURNAL
(2023)
Article
Computer Science, Artificial Intelligence
Guofa Li, Zefeng Ji, Shen Li, Xiao Luo, Xingda Qu
Summary: This paper proposes a new off-policy imitation learning method for autonomous driving by using task knowledge distillation, which overcomes the dependence on large scales of time-consuming, laborious, and reliable labels in existing behavioral cloning methods. The experiment results show that our method can achieve satisfactory route-following performance in realistic urban driving scenes and can transfer the driving strategies to new unknown scenes under various illumination and weather scenarios for autonomous driving.
IEEE TRANSACTIONS ON INTELLIGENT VEHICLES
(2023)
Article
Computer Science, Artificial Intelligence
Guofa Li, Yifan Qiu, Yifan Yang, Zhenning Li, Shen Li, Wenbo Chu, Paul Green, Shengbo Eben Li
Summary: End-to-end approaches are a promising solution for AV decision-making, but their deployment is often hindered by high computational burden. To address this, we propose a lightweight transformer-based end-to-end model with risk awareness for AV decision-making. We introduce a lightweight network with depth-wise separable convolution and transformer modules to extract image semantics from trajectory data. We then assess driving risk using a probabilistic model with position uncertainty and integrate it into deep reinforcement learning to find strategies with minimum expected risk. The proposed method is evaluated in three lane change scenarios to validate its superiority.
IEEE TRANSACTIONS ON INTELLIGENT VEHICLES
(2023)
Article
Engineering, Electrical & Electronic
Guofa Li, Xingyu Chi, Xingda Qu
Summary: This paper proposes an improved bidirectional feature pyramid module (BiFPN) and a channel attention module (Seblock) to address issues in existing methods based on monocular camera sensor. The improved BiFPN facilitates efficient fusion of multi-scale features, while Seblock enhances useful information by redistributing channel feature weights. Experimental results demonstrate that this method is competitive with state-of-the-art algorithms and preserves fine-grained texture of scene depth.
AUTOMOTIVE INNOVATION
(2023)
Article
Engineering, Electrical & Electronic
Guofa Li, Delin Ouyang, Xin Chen, Wenbo Chu, Bing Lu, Caizhi Zhang, Xiaolin Tang, Gang Guo
Summary: This article introduces a reinforced one-shot multiobject tracking system that utilizes new methods to improve tracking accuracy. The article first presents a receptive field module to address the problem of target scale transformation, then designs an attention mechanism network to extract channel and positional information, and finally combines circle loss with Euclidean distance optimization and cross-entropy loss to enhance the learning of discriminative embeddings.
IEEE SENSORS JOURNAL
(2023)
Article
Engineering, Electrical & Electronic
Guofa Li, Junda Li, Cong Wang, Qianlei Peng, Caizhi Zhang, Feng Gao, Xiaolin Tang, Gang Guo
Summary: In the development of autonomous driving technologies, 3-D car detection based on LiDAR has been recognized as a key research topic. However, LiDAR-based methods are limited by the incomplete point cloud problem caused by signal missing and object occlusion. To tackle this challenge, we introduce a novel early fusion method called key supplement, which supplements the key area with pseudo point cloud generated by images to address the incomplete point cloud problem. We also propose a multimodal method to reconstruct the incomplete point cloud. Experimental results demonstrate that our proposed key supplement method improves the performances of state-of-the-art LiDAR-based methods.
IEEE SENSORS JOURNAL
(2023)
Article
Engineering, Civil
Guofa Li, Xuanhu Qian, Xingda Qu
Summary: High-quality fusion images with infrared and visible information play a crucial role in intelligent and safe driving. To address the issue of unclear fusion images due to noise in the infrared image and loss of texture information from the visible image, we propose a novel two-stage network called SOSMaskFuse. The experimental results demonstrate that our proposed network outperforms other algorithms in terms of quality and effectiveness.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2023)
Article
Engineering, Civil
Qingkun Li, Yizi Su, Wenjun Wang, Zhenyuan Wang, Jibo He, Guofa Li, Chao Zeng, Bo Cheng
Summary: Although latent hazard notification for highly automated driving is expected to enhance traffic safety, its practical effects have yet to be verified. This study investigated the expected safety benefits and driver behavioral adaptation based on structural equation modeling. The findings reveal that while latent hazard notification significantly improves driver attention and enhances traffic safety, it also increases driver trust and impairs traffic safety due to driver behavioral adaptation.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Wenbo Li, Yaodong Cui, Yintao Ma, Xingxin Chen, Guofa Li, Guanzhong Zeng, Gang Guo, Dongpu Cao
Summary: This article introduces a new dataset, the driver emotion facial expression (DEFE) dataset, for analyzing drivers' spontaneous emotions. The dataset consists of facial expression recordings of 60 participants during driving. Each participant completed driving tasks in the same scenario and rated their emotional responses using the dimensional emotion method and discrete emotion method. Classification experiments were conducted to recognize arousal, valence, dominance scales, emotion category, and intensity as baseline results. Additionally, the article compares emotion recognition results between dynamic driving and static life scenarios through facial expressions.
IEEE TRANSACTIONS ON AFFECTIVE COMPUTING
(2023)
