检索结果-内蒙古大学图书馆

49th IEEE International conference on Acoustics, Speech, and Signal processing (ICASSP)

作者： Wu, Kexin Tang, Fan Liu, Ning Deussen, Oliver Le, Thi-Ngoc-Hanh Dong, Weiming Lee, Tong-Yee Jilin Univ Jilin Jilin Peoples R China ICT CAS Beijing Peoples R China Midea Grp Beijiaozhen Peoples R China Univ Konstanz Constance Germany Natl Cheng Kung Univ Tainan Taiwan CASIA Beijing Peoples R China

ISBN: (纸本)9798350344868;9798350344851

Deploying style transfer methods on resource-constrained devices is challenging, which limits their real-world applicability. To tackle this issue, we propose using pruning techniques to accelerate various visual style transfer methods. We argue that typical pruning methods may not be well-suited for style transfer methods and present an iterative correlation-based channel pruning (ICCP) strategy for encoder-transform-decoder-based image/video style transfer models. The correlation-based channel regularization preserves the feature distributions for content and style references, and the iterative pruning strategy prevents layer collapse when pruning on the encoder-decoder structure. Experiments demonstrate that the proposed ICCP can generate visual competitive results compared to SOTA style transfer methods and significantly reduces the number of parameters (at least 70K) and inference time. Model is available at https://***/wukx-wukx/ICCP.

关键词： visual style transfer model pruning

来源：评论

学校读者我要写书评

暂无评论

Interpretable Neural Networks for video Separation: Deep Unfolding RPCA With Foreground Masking

引用

IEEE TRANSACTIONS ON image processing 2024年 33卷 108-122页

作者： Joukovsky, Boris Eldar, Yonina C. Deligiannis, Nikos Vrije Univ Brussel Dept Elect & Informat B-1050 Ixelles Belgium Weizmann Inst Sci IL-7610001 Rehovot Israel

We present two deep unfolding neural networks for the simultaneous tasks of background subtraction and foreground detection in video. Unlike conventional neural networks based on deep feature extraction, we incorporate domain knowledge models by considering a masked variation of the robust principal component analysis problem (RPCA). With this approach, we separate video clips into low-rank and sparse components, respectively corresponding to the backgrounds and foreground masks indicating the presence of moving objects. Our models, coined ROMAN-S and ROMAN-R, map the iterations of two alternating direction of multipliers methods (ADMM) to trainable convolutional layers, and the proximal operators are mapped to non-linear activation functions with trainable thresholds. This approach leads to lightweight networks with enhanced interpretability that can be trained on limited data. In ROMAN-S, the correlation in time of successive binary masks is controlled with side-information based on l(1)-l(1) minimization. ROMAN-R enhances the foreground detection by learning a dictionary of atoms to represent the moving foreground in a high-dimensional feature space and by using reweighted-l(1)-l(1) minimization. Experiments are conducted on both synthetic and real video datasets, for which we also include an analysis of the generalization to unseen clips. Comparisons are made with existing deep unfolding RPCA neural networks, which do not use a mask formulation for the foreground, and with a 3D U-Net baseline. Results show that our proposed models outperform other deep unfolding networks, as well as the untrained optimization algorithms. ROMAN-R, in particular, is competitive with the U-Net baseline for foreground detection, with the additional advantage of providing video backgrounds and requiring substantially fewer training parameters and smaller training sets.

关键词： Deep learning deep unfolding masked RPCA video separation foreground detection

来源：评论

学校读者我要写书评

暂无评论

A quantum moving target segmentation algorithm based on mean background modeling

引用

QUANTUM INFORMATION processing 2024年第11期23卷 1-20页

作者： Wang, Lu Liu, Yuxiang Meng, Fanxu Zhang, Zaichen Yu, Xutao Southeast Univ Sch Informat Sci & Engn 2 Southeast Univ Rd Nanjing 211189 Jiangsu Peoples R China Southeast Univ State Key Lab Millimeter Waves 2 Southeast Univ Rd Nanjing 211189 Jiangsu Peoples R China Southeast Univ Frontiers Sci Ctr Mobile Informat Commun & Secur 2 Southeast Univ Rd Nanjing 211189 Jiangsu Peoples R China Nanjing Tech Univ Coll Artificial Intelligence 30 Puzhu Nan Rd Nanjing 211800 Jiangsu Peoples R China Southeast Univ Natl Mobile Commun Res Lab 2 Southeast Univ Rd Nanjing 211189 Jiangsu Peoples R China Purple Mt Labs 9 Mozhou Dong Rd Nanjing 211111 Jiangsu Peoples R China

Classical algorithms for moving target segmentation have made significant progress, but the real-time problem has become a significant obstacle for them as the data volume grows. Quantum computing has been proven to be beneficial for image segmentation, but is still scarce for video. In this paper, a quantum moving target segmentation algorithm based on mean background modeling is proposed, which can utilize the quantum mechanism to do segmentation operations on all pixels in a video at the same time. In addition, a quantum divider with lower quantum cost is designed calculate pixel mean, and then, a number of quantum modules are designed according to the algorithmic steps to build the complete quantum algorithmic circuit. For a video containing 2m\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2<^>m$$\end{document} frames (every frame is a 2nx2n\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2<^>n \times 2<^>n$$\end{document} image with q grayscale levels), the proposed algorithm is superior compared to both existing quantum and classical algorithms. Finally, the experiment on IBM Q shows the feasibility of the algorithm in the NISQ era.

关键词： Quantum image processing Quantum video representation Quantum video segmentation Mean background modeling

来源：评论

学校读者我要写书评

暂无评论

AI-Powered Multi View Face video Super-Resolution Techniques for real-time video processing 2

AI-Powered Multi View Face Video Super-Resolution Techniques...

引用

2nd IEEE International conference on Integrated Intelligence and Communication Systems, ICIICS 2024

作者： Manjunatha, D.V. Maindola, Meenakshi Jose, Reny Kaliappan, S. Patel, Mitul Maranan, Ramya Department of Electronics & Communication Engineering Alvas Institute of Engineering & Technology Moodabidri India Department of Computer Science & Engineering Graphic Era Deemed to be University Dehradun India Idukki India Division of Research and Development Lovely Professional University Phagwara India Faculty of Engineering and Technology Parul institute of Engineering and Technology Parul University Vadodara India Department of Research and Innovation Saveetha School of Engineering SIMATS Chennai India

ISBN: (纸本)9798331504960

In order to improve the visual quality of low-resolution video frames, this study introduces a new superresolution method for real-time video processing that is powered by artificial intelligence. With little computational overhead, the suggested approach uses a hybrid deep learning model to reconstruct high-resolution frames. This model mixes Convolutional Neural Networks (CNNs) with Generative Adversarial Networks (GANs). The suggested method outperforms state-of-the-art methods by 12.3% in PSNR and 8.7% in SSIM, according to experimental results on benchmark video datasets. The PSNR is 34.56 dB and the SSIM is 0.94. Also, running on a GTX 1080Ti GPU, the system shows an average processing speed of 30 fps, which is great for real-time apps. The suggested strategy is effective in eliminating artifacts, improving visual clarity, and retaining fine details in various real-time video processing contexts, according to both quantitative and qualitative results. © 2024 IEEE.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Proceedings of the 2020 4th International conference on video and image processing, ICVIP 2020

Proceedings of the 2020 4th International Conference on Vide...

引用

4th International conference on video and image processing, ICVIP 2020

ISBN: (纸本)9781450389075

The proceedings contain 39 papers. The topics discussed include: optimization method of loop detection based on shadow compensation;real time lane detection model based on lightweight;research on image detection algorithm based on improved retinanet;a study of student learning status classification based on the detection of key objects within the visual field;an outlier detection method based on symmetry and curvature threshold;research on adaptive object detection method of kernel correlation filtering;attention enhanced multi-patch deformable network for image deblurring;recaptured image forensics based on image illumination and texture features;and using temporal convolutional networks to enable action recognition for construction equipment.

关键词：

来源：评论

学校读者我要写书评

暂无评论

realization of a real-time image Denoising System for Dashboard Camera Applications

引用

IEEE TRANSACTIONS ON CONSUMER ELECTRONICS 2022年第2期68卷 181-190页

作者： Yu, Chu Hou, Li-Zhong Natl Ilan Univ Dept Elect Engn Yilan 260007 Taiwan

Noise interference during the acquisition of digital images can severely degrade image quality, particularly for images captured under low-light conditions;however, the removal of image noise requires sophisticated digital image processing systems. This study presents a hardware-based solution to real-time image denoising using an existing algorithm designed for the removal of mixed impulse noise (salt-and-pepper and random-valued impulse noise), while preserving image edge details and image borders, without the need for additional computation time or memory capacity. Note that mixed impulse noise is typical of most real-world situations, such as the video noise associated with dashboard cameras. The proposed design was implemented using 180 nm complementary metal-oxide-semiconductor (CMOS) technology, consuming only 21.7 mW when operated at 200 MHz. This operating frequency allows the proposed chip to process noisy video streams with resolution of 1920x1080 at 60 frames per second in real time. In terms of image restoration, the proposed algorithm achieved image quality on par with that achieved using software simulation. We also demonstrated the efficacy of the proposed scheme in denoising noisy video images from a dashboard camera.

关键词： Digital images impulse noise dashboard cameras

来源：评论

学校读者我要写书评

暂无评论

Enabling real-time video Analytics with Adaptive Sampling and Detection-based Tracking in Edge Computing

Enabling Real-time Video Analytics with Adaptive Sampling an...

引用

IEEE conference on Global Communications (IEEE GLOBECOM) - Intelligent Communications for Shared Prosperity

作者： Wang, Yilan Liu, Zhicheng Zhao, Yunfeng Wang, Xiaofei Qiu, Chao Tianjin Univ Coll Intelligence & Comp Tianjin Peoples R China

ISBN: (纸本)9798350310900

With the popularization of visual machine learning, intelligent video analytics can automatically analyze and extract information from video streams, yet it brings heavy computing burdens. Edge computing can improve the processing experience by bringing computing resources near users. On top of this, various processing methods and settings have different resource requirements and output different user experiences. How to dynamically select the video processing configuration according to system states becomes a critical problem that remains to be addressed. In this paper, we propose an edge-assisted video analytic framework based on adaptive sampling and detection-based tracking. We design four functional modules to realize a cooperative computing processing flow. We consider two performance metrics, recognition accuracy and processing time to estimate the experience of real-time video analytics. Further, we design an online configuration method based on Double Deep Q-Network, which can adaptively select analytic configurations under the condition of system dynamics. Experimental results based on a real dataset demonstrate the superior performance of the proposed framework on reward, mean Intersection over Union (IoU), and processing time.

关键词： video analytics object detection edge computing

来源：评论

学校读者我要写书评

暂无评论

Visualization simulation of aerobic gymnastics movements using thermal radiation images based on image tracking: Mechanism of heat transfer during exercise

引用

THERMAL SCIENCE AND ENGINEERING PROGRESS 2025年 58卷

作者： Bin, Dongsong Wang, Yong Guangxi Univ Foreign Languages Phys Educ Dept Nanning 530222 Guangxi Peoples R China GuangXi MinZu Univ Sch Phys Educ & Hlth Sci Nanning 530006 Guangxi Peoples R China

In the field of sports science, visualization techniques are being used more and more widely, especially in the analysis of athletes' movements and body functions. Traditional motion analysis methods often rely on video recording and sensor data, but these methods are in the capture of subtle physiological changes, this study developed a thermal radiation image analysis method based on image tracking for visual simulation of aerobics movements. In this paper, a high-resolution thermal camera is used to capture the thermal radiation images of aerobics athletes when they perform specific actions in real time, and image processing software is used to analyze the acquired thermal radiation images and extract the key temperature data and thermal radiation patterns. Combined with the video data recorded by the high-speed camera, the athletes' movements are accurately tracked to ensure that the thermal radiation data corresponds to the specific movements. Statistical methods were used to analyze the thermal radiation image data to explore the influence of different movements on the body's thermal radiation distribution. According to the analysis results, a visual model is constructed to simulate the thermal radiation change process of athletes performing *** results show that there are significant differences in the distribution of heat radiation in different parts of the body when aerobics athletes perform different movements. When performing high-intensity jumping and rotating movements, increased muscle activity leads to increased local temperature;In static stretching, the muscles relax and the local temperature is relatively low. These changes were clearly observed through visual simulation of thermal radiation images, and it was found that the distribution of thermal radiation was closely related to the intensity of muscle activity and blood circulation.

关键词： image tracking Thermal radiation image Aerobics Visual simulation of action Motion heat transfer mechanism

来源：评论

学校读者我要写书评

暂无评论

video Analytics with FPGA based smart cameras for Object Recognition in the game of field hockey 1

Video Analytics with FPGA based smart cameras for Object Rec...

引用

1st IEEE International conference on Networking and Communications, ICNWC 2023

作者： Praveena, M. Vinoth Thyagarajan, V. Venkatasubramani, V.R. Thiagarajar College of Engineering Dept of ECE Tamilnadu Madurai India

ISBN: (纸本)9798350336009

video Analytics is an image processing method that takes video as input and extracts information. It is the latest technology that analyzes and processes a digital video signal for real time monitoring. Some of the countries of south and eastern hemisphere are top at field hockey because they are very much developed in technology. Even though, field hockey is our national game we are not in the top because of lack of technology. By using video analytics, we can analyze the player's video and from that they can improve their performance. Not only performance analysis there are many advantages like weapon detection, missing person in stadium, replay of players, real time instant videos, scores, reconstruct story of a match, violent detection can be analyzed. The experimental setup for this work is camera with SD-card, VGA interfacing with Field Programmable Gate Array (FPGA). The SD-card communicates with FPGA using SD-card interface. image datas stored in SD-card are read by FPGA and stored in its memory. Then that image is processed using Convolutional Neural Network (CNN). This algorithm detects the objects in an image. Then object recognition is performed for videos to make decisions. This whole processing of image is performed inside FPGA. Finally, to display the processed image in monitor, FPGA communicates with VGA interface through VGA cable. real time videos will be displayed by increasing the frame rate. Our work is implemented in Altera DE 1 FPGA board of device EP2C20F484C7N which comes under family of cyclone II and image displayed on monitor with resolution and frequency of 1366x768-60 Hz. © 2023 IEEE.

关键词： Cameras

来源：评论

学校读者我要写书评

暂无评论

Depth-Aware Weather-Enhanced 3D Object Detection for Roadside Infrastructure 8

Depth-Aware Weather-Enhanced 3D Object Detection for Roadsid...

引用

8th International conference on video and image processing, ICVIP 2024

作者： Li, Yanfei Zeng, Yuan Gong, Yi Department of Electronics and Electrical Engineering Southern University of Science and Technology China College of Big Data and Internet Shenzhen Technology University China

ISBN: (数字)9781510689244

ISBN: (纸本)9781510689237

Monocular 3D object detection is a crucial topic in autonomous driving and Intelligent transportation systems (ITS). Most existing methods are evaluated on clean datasets but exhibit arresting performance degradation in the real world with varying weather conditions. image enhancement methods have been introduced in existing vehicle-centric works to address the impact of diverse weather on detection performance and improve the robustness of the model. Unlike vehicle-centric systems, roadside infrastructure can enlarge perception range and is less affected by vehicle occlusion to increase the response time in case of danger. To enhance the robustness of detection models on roadside infrastructure, this work introduces an image enhancement method incorporating depth information in modeling climate changes. We use a classification network trained on real weather data to validate the effectiveness of the enhancement and use state-of-the-art roadside infrastructure monocular detection models to evaluate the effectiveness of our method on object detection. Extensive experiments demonstrate that our method can enhance the robustness and generalization of detection models under various weather conditions. © 2025 SPIE.

关键词： Object detection Rain image enhancement Data modeling Education and training Cameras 3D modeling Adverse weather Performance modeling image contrast enhancement

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：