Article
Engineering, Electrical & Electronic
Md. Moniruzzaman, Zhaozheng Yin
Summary: This paper investigates the problem of weakly supervised temporal action localization and proposes a novel collaborative network to accurately localize action instances in untrimmed videos with only video-level supervision. The network models foreground, background, and action separately and collaboratively, reducing false positives and false negatives.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Computer Science, Information Systems
M. Ozan Tezcan, Prakash Ishwar, Janusz Konrad
Summary: Background subtraction (BGS) is a crucial video processing task with deep learning-based supervised algorithms showing strong performance, although their performance drops significantly when applied to unseen videos. This study introduces spatio-temporal data augmentations to enhance the leading video-agnostic BGS algorithm BSUV-Net, resulting in a significant improvement in performance on the CDNet-2014 dataset.
Article
Computer Science, Artificial Intelligence
Hongrui Zhang, Huan Li
Summary: This paper proposes an interactive spatio-temporal feature learning network (ISFLN) for video foreground detection (VFD), which utilizes deep and shallow spatio-temporal information extraction, multi-scale feature extraction, and multi-level feature enhancement to achieve more effective foreground detection.
COMPLEX & INTELLIGENT SYSTEMS
(2022)
Article
Mathematics, Applied
Jun-Hao Zhuang, Yi-Si Luo, Xi-Le Zhao, Tai-Xiang Jiang, Yi Chang, Jun Liu
Summary: Outdoor video rain streaks removal is a crucial problem in video processing. Traditional methods based on prior knowledge are insufficient for capturing complex structures of real-world videos, while deep learning methods with large model capacities show promise but require abundant training data. We propose an unsupervised method for video rain streaks removal that solely utilizes rainy videos, combining deep neural networks to capture foreground and background components and incorporating total variation regularization for structured rain streaks.
JOURNAL OF COMPUTATIONAL AND APPLIED MATHEMATICS
(2024)
Article
Computer Science, Artificial Intelligence
Yumei Sun, Chuanming Tang, Hui Luo, Qingqing Li, Xiaoming Peng, Jianlin Zhang, Meihui Li, Yuxing Wei
Summary: In this paper, a spatio-temporal joint-modeling tracker named STTrack is proposed for continuous object tracking in video sequences. The proposed method concentrates on the temporal connection of the object by employing a time-sequence iteration strategy (TSIS) and captures the spatio-temporal correlation of the object between frames using a novel spatial temporal interaction Transformer network (STIN). Experimental results demonstrate that STTrack achieves excellent performance on six tracking benchmark datasets while operating in real-time.
KNOWLEDGE-BASED SYSTEMS
(2024)
Article
Thermodynamics
Zhibin Niu, Junqi Wu, Xiufeng Liu, Lizhen Huang, Per Sieverts Nielsen
Summary: This paper proposes a novel energy demand-side management approach based on smart meter data, utilizing spatio-temporal visual analysis to discover urban energy consumption patterns, identify energy-saving potentials, plan energy supply, and improve energy efficiency. Through empirical studies, five typical energy consumption patterns and demand shift patterns are identified in the Pudong district.
Article
Engineering, Electrical & Electronic
Subhaluxmi Sahoo, Pradipta Kumar Nanda
Summary: Moving cast shadow identification and extraction are common problems in visual surveillance applications. This paper proposes a new scheme to detect moving objects in video frames while removing shadows. The scheme utilizes background modeling and model learning, with weights adaptively determined for probabilistic fusion in the feature space. Online background model learning and elimination of residual shadows are also implemented. Experimental results demonstrate improved performance compared to existing methods.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Article
Computer Science, Information Systems
Jae-Yeul Kim, Jong-Eun Ha
Summary: This paper proposes an algorithm that improves the performance of foreground object detection in visual surveillance in unseen environments by integrating spatial and temporal information. The algorithm utilizes a spatio-temporal fusion network (STFN) to extract temporal and spatial information and employs a semi-foreground map for stable training.
Article
Engineering, Electrical & Electronic
Yi Zhong, Guoqiang Mao, Xiaohu Ge, Fu-Chun Zheng
Summary: This paper introduces the concept of massive and sporadic access (MSA) to describe the massive access of IoT devices, evaluating the temporal correlation of interference and successful transmission events. An approximation is proposed to address the complexity of interactions among queues, assuming all nodes move so fast that their locations are independent at different time slots.
IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Yuming Fang, Zhaoqian Li, Jiebin Yan, Xiangjie Sui, Hantao Liu
Summary: Video quality assessment has been widely discussed, where most models employ RNNs to capture temporal quality variation. However, it is unclear whether RNNs learn spatio-temporal representation or just redundantly aggregate spatial features. In this study, we train VQA models with carefully designed frame sampling and fusion methods, leading to the finding that RNNs do not facilitate quality-aware spatio-temporal feature learning and sparsely sampled frames can achieve competitive performance. This is the first work to explore spatio-temporal modeling in VQA.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2023)
Article
Green & Sustainable Science & Technology
Vito Imbrenda, Rosa Coluzzi, Valerio Di Stefano, Gianluca Egidi, Luca Salvati, Caterina Samela, Tiziana Simoniello, Maria Lanfredi
Summary: The study examines the long-term equilibrium conditions of desertification processes in Italy and finds evidence of path dependence. The results show a traditional divergence in desertification risk between Southern and Northern Italy in the first period, while the second period saw convergence in some southern areas and divergence in northern areas, leading to spatial homogenization towards higher vulnerability levels. The importance of spatially explicit approaches for effective policy strategies in addressing land vulnerability to degradation in Italy is highlighted.
Article
Automation & Control Systems
Yu Zhou, Han-Xiong Li, Sheng-Li Xie
Summary: This article proposes a systematic approach for fast modeling of the distributed battery thermal process by adopting time/space separation and incremental learning. It achieves high efficiency and flexibility in reconstructing desired temperature distribution.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2022)
Article
Green & Sustainable Science & Technology
Patricia Maldonado-Salguero, Maria Carmen Bueso-Sanchez, Angel Molina-Garcia, Juan Miguel Sanchez-Lozano
Summary: This study proposes a methodology to characterize and cluster spatio-temporal solar resource variability through global horizontal irradiance analysis. The study selected Spain as a case study and evaluated spatial variability and geographical clustering differences of solar resources in different time windows. The research findings contribute to understanding the spatiotemporal variations of solar resources in Spain.
Article
Computer Science, Artificial Intelligence
Xiaohan Lin, Xiaolong Zou, Zilong Ji, Tiejun Huang, Si Wu, Yuanyuan Mi
Summary: The study introduces a novel brain-inspired computational model for generic spatio-temporal pattern recognition, consisting of reservoir and decision-making modules, capable of extracting frequency and order information of temporal inputs and performing well in looming pattern discrimination and gait recognition tasks, outperforming deep learning counterparts with limited training data.
Article
Computer Science, Artificial Intelligence
Jiadong Yan, Yuzhong Chen, Zhenxiang Xiao, Shu Zhang, Mingxin Jiang, Tianqi Wang, Tuo Zhang, Jinglei Lv, Benjamin Becker, Rong Zhang, Dajiang Zhu, Junwei Han, Dezhong Yao, Keith M. Kendrick, Tianming Liu, Xi Jiang
Summary: In this study, a novel Multi-Head Guided Attention Graph Neural Network (Multi-Head GAGNN) was proposed to model both spatial and temporal patterns of holistic functional brain networks. The results showed that the Multi-Head GAGNN outperformed other state-of-the-art models in modeling brain function and predicting cognitive behavioral measures in individuals.
MEDICAL IMAGE ANALYSIS
(2022)
Article
Engineering, Electrical & Electronic
Subrata Kumar Mohanty, Suvendu Rup, M. N. S. Swamy
Summary: An efficient foreground detection scheme is proposed which combines texture features and color/gray value features, computes similarity distance measure using Canberra distance, and adaptively selects threshold value under dynamic background conditions, outperforming existing schemes in terms of quantitative and qualitative measures.
DIGITAL SIGNAL PROCESSING
(2021)