Article
Environmental Sciences
Qin Zhang, Jialang Xu, Matthew Crane, Chunbo Luo
Summary: Wind sensing by learning from video clips can improve the spatiotemporal resolution of weather records at the city scale. This paper introduces a novel approach to train cameras to sense the wind by capturing motion information using optical flow and machine learning models. A dataset of labeled video clips covering eleven wind classes is used to evaluate the proposed method, achieving an accuracy of 86.69%.
SCIENCE OF THE TOTAL ENVIRONMENT
(2022)
Article
Engineering, Electrical & Electronic
Weiqing Yan, Yiqiu Sun, Wujie Zhou, Zhaowei Liu, Runmin Cong
Summary: This study proposes an unsupervised deep video stabilization method that addresses the influence of moving objects on video stabilization through robust homography estimation. The method estimates a foreground mask as a preprocessing step, improves motion alignment through low-level confidence feature extraction, and captures spatial correspondence between frames through local and global feature extraction.
IEEE SIGNAL PROCESSING LETTERS
(2023)
Article
Computer Science, Artificial Intelligence
Le Wang, Hongzhen Liu, Sanping Zhou, Wei Tang, Gang Hua
Summary: Video panoptic segmentation is a challenging task in computer vision that requires both frame-level segmentation and instance association across frames. To address the issues of identity switches and ambiguous segmentation boundaries, we propose a simple yet effective Instance Motion Tendency Network (IMTNet) that utilizes a global motion tendency map and a hierarchical classifier for motion boundary refinement.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2023)
Article
Computer Science, Information Systems
Fen Xiao, Huiyu Luo, Wenlei Zhang, Zhen Li, Xieping Gao
Summary: The paper proposes a spatial and motion dual-stream framework for video saliency detection, which effectively describes and utilizes the motion information contained in video data and models the inter-frame correlation using the convolutional gated recurrent unit (convGRU).
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Engineering, Electrical & Electronic
Chao Xu, Jiangning Zhang, Mengmeng Wang, Guanzhong Tian, Yong Liu
Summary: This paper proposes a Multi-level Spatial-Temporal (MST) feature aggregation framework to address the challenges in video object detection. The framework fully exploits spatial-temporal features at frame level, proposal level, and instance level, and introduces a Deformable Feature Alignment (DAlign) module to improve pixel-level spatial alignment for feature aggregation.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Article
Computer Science, Artificial Intelligence
Fan Yang, Li Su, Jinxiu Zhao, Xuena Chen, Xiangyu Wang, Na Jiang, Quan Hu
Summary: Inspired by biological vision mechanism, event-based cameras have been developed to overcome the limitations of traditional frame-based cameras by capturing continuous object motion and detecting brightness changes independently and asynchronously. Complementarily, spiking neural networks (SNNs) offer asynchronous computations and exploit the sparsity of spatio-temporal events. However, event-based pixel-wise optical flow estimations face challenges in generating dense scene information due to sparse and uneven event camera outputs, as well as poor moving objects tracking caused by local receptive fields of neural networks. To address these issues, an improved event-based self-attention optical flow estimation network (SA-FlowNet) is proposed, which utilizes cross-domain and temporal self-attention mechanisms to capture long-range dependencies and extract temporal and spatial features from event streams efficiently.
IET COMPUTER VISION
(2023)
Article
Neurosciences
XiaoLe Liu, Si-yang Yu, Nico A. Flierman, Sebastian Loyola, Maarten Kamermans, Tycho M. Hoogland, Chris I. De Zeeuw
Summary: OptiFlex is a novel multi-frame animal pose estimation framework that integrates a flexible base model and an OpticalFlow model to improve prediction accuracy by considering variability in animal body shape and incorporating temporal context.
FRONTIERS IN CELLULAR NEUROSCIENCE
(2021)
Article
Computer Science, Information Systems
Kaiqiao Wang, Peng Liu
Summary: This paper proposes a novel approach based on the concept of local compensation to address the challenges in video frame interpolation. By introducing comprehensive contextual feature extraction and motion-guided feature fusion modules, more refined optical flow estimation can be achieved, leading to higher-quality results.
Article
Geochemistry & Geophysics
Jialian Wu, Xin Su, Qiangqiang Yuan, Huanfeng Shen, Liangpei Zhang
Summary: The research introduces a novel multi-object tracking method in satellite video, SF and motion feature-guided multi-object tracking (SFMFMOT), which utilizes slow features and motion features for continuous tracking of moving vehicles. By incorporating bounding box proposals and optimization strategies, the method successfully reduces false alarms, improves recall rate, and demonstrates superior tracking performance in satellite videos.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Engineering, Electrical & Electronic
Bo Yan, Weimin Tan, Chuming Lin, Liquan Shen
Summary: A novel fine-grained motion estimation approach (FGME) for video frame interpolation is proposed, achieving high-quality intermediate frame synthesis through multi-scale optimization and multiple motion features estimation. The simple fully convolutional neural network with three refinement scales and four motion features produces state-of-the-art results.
IEEE TRANSACTIONS ON BROADCASTING
(2021)
Article
Engineering, Electrical & Electronic
Thangaswamy Judi Vennila, Vanniappan Balamurugan
Summary: MultiHuman Tracking (MHT) is an important research area in video surveillance applications. This article proposes a rough set framework to address the challenges of tracking human objects in consecutive frames, including the spatial disorder, nonlinear motion, and occlusion. The framework utilizes a modified bounding box generation technique and a rough set classifier to identify human objects and achieved better detection and tracking accuracy compared to existing algorithms.
IEEE SENSORS JOURNAL
(2023)
Article
Computer Science, Artificial Intelligence
Xinqian Gu, Hong Chang, Bingpeng Ma, Shiguang Shan
Summary: The proposed Motion Feature Aggregation (MFA) method efficiently models and aggregates motion information in the feature map level for video-based re-identification (re-id), consisting of coarse-grained and fine-grained motion learning modules that can model motion information from different granularities and are complementary to each other.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Environmental Sciences
Yan Zhang, Deng Chen, Yuhui Zheng
Summary: Video satellites provide a novel method for real-time observation of dynamic changes in ground targets. To address the challenges of target tracking in satellite videos, a multi-feature correlation filter with motion estimation is proposed. Experimental results demonstrate that the algorithm achieves high tracking performance.
Article
Engineering, Electrical & Electronic
Shihao Zou, Yuanlu Xu, Chao Li, Lingni Ma, Li Cheng, Minh Vo
Summary: This paper proposes Snipper, a unified framework that performs multi-person 3D pose estimation, tracking, and motion forecasting simultaneously in a single stage. By employing an efficient and powerful deformable attention mechanism, Snipper aggregates spatiotemporal information from the video snippet and uses a video transformer to encode and decode pose features for multi-person pose queries. The experimental results demonstrate the effectiveness of Snipper on three challenging public datasets, where it achieves comparable performance with specialized state-of-the-art models for pose estimation, tracking, and forecasting.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Computer Science, Artificial Intelligence
Qing Li, Cheng Wang, Xin Li, Chenglu Wen
Summary: The study introduces a novel neural network architecture, FeatFlow, for estimating 3D motions from unstructured point clouds, with good generalization performance in scenarios with limited sensor data.
PATTERN RECOGNITION
(2021)