检索结果-内蒙古大学图书馆

Robust Multi-Object Tracking Using Vision Sensor with Fine-Grained Cues in Occluded and Dynamic Scenes

IEEE Sensors Journal 2025年第11期25卷 20547-20560页

作者： Hu, Yaoqi Sun, Jinqiu Jin, Hao Niu, Axi Yan, Qingsen Zhu, Yu Zhang, Yanning School of Computer Science China School of Astronautics China National Engineering Laboratory for Integrated Aero-Space-Ground-Ocean Big Data Application Technology Northwestern Polytechnical University Dongxiang Road Shaanxi Xi'an710129 China

Multi-object tracking (MOT) using vision sensors remains a challenging problem, particularly in dynamic backgrounds and severe occlusions. Existing methods, relying on holistic appearance or spatial cues, fail to capture detailed information within object regions and the background, resulting in inaccurate and inconsistent tracking. To address these issues, we propose a novel Detail-Driven Multi-Object Tracking (DD-MOT) method that revisits and leverages fine-grained cues to discover the dynamics of both object regions and the background, facilitating the recovery and association of trajectories. Specifically, the proposed method consists of three key modules: (i) Points Trajectories Generator, (ii) Camera Motion and Occlusion Compensation, and (iii) Fine- and Coarsegrained Association. The Points Trajectories Generator is responsible for generating fine-grained cues by sampling an initial set of points, generating point trajectories, and refining the initial set of points. The Camera Motion and Occlusion Compensation module utilizes the background and object point trajectories to correct background motion and recover occluded object bounding boxes. Finally, the Fine- and Coarse-grained Association module leverages point trajectory cues to assist in establishing a more effective association strategy, combining both fine-grained and coarse-grained spatial cues from the object bounding boxes. We evaluate our DD-MOT method on several benchmark datasets, including MOT17, MOT20, demonstrating that it consistently outperforms state-of-the-art (SOTA) methods in key metrics such as HOTA (64.2, 62.6), MOTA (80.8, 78.1), and IDF1 (78.8, 77.1). © 2025 IEEE.

关键词： Motion compensation

来源：评论

学校读者我要写书评

暂无评论

Mental illness detection using sentiment analysis in social media 9th

Mental illness detection using sentiment analysis in social ...

引用

9th International Conference on computer Science and Computational Intelligence, ICCSCI 2024

作者： Odja, Kasimirus Derryl Widiarta, Jasson Purwanto, Eko Setyo Ario, Muhamad Keenan Computer Science Department School of Computer Science Bina Nusantara University Jakarta11480 Indonesia Mobile Application & Technology Program Computer Science Department School of Computer Science Bina Nusantara University Jakarta11480 Indonesia

This research tries to detect mental illness using sentiment analysis on Reddit data, as well as comparing the performance of the k-Nearest Neighbors (k-NN), Random Forest, and Neural Network models. Using text post data from the pre-pandemic and post-pandemic periods, we concluded that the Random Forest model had the highest overall performance with an F1 Score, accuracy, recall and precision of 80.6%, making it quite effective in detecting depression. Even though the Neural Network model shows slightly lower accuracy, namely 79%, in fact this model has the lowest error rate, namely 0.06496. The k-NN model showed the lowest accuracy and higher error rate. These findings highlight the potential of sentiment analysis and machine learning in identifying mental health issues on social media and suggest that better models can improve early detection and intervention efforts. © 2024 The Authors.

关键词： Neural network models

来源：评论

学校读者我要写书评

暂无评论

The future of virtual reality: Prospect and problems 9th

The future of virtual reality: Prospect and problems

引用

9th International Conference on computer Science and Computational Intelligence, ICCSCI 2024

作者： Lie, Andrew Alfonso Buntoro, Owen Tamashi Purwanto, Eko Setyo Ario, Muhamad Keenan Computer Science Department School of Computer Science Bina Nusantara University Jakarta11480 Indonesia Mobile Application & Technology Program Computer Science Department School of Computer Science Bina Nusantara University Jakarta11480 Indonesia

Virtual Reality (VR) is a technology that allows users to interact with a simulated environment created by virtual reality capabilities through the intermediary of computers, simulator tools, and others, providing new experiences and can feel experiences that are not possible in the real world. This research aims to identify and analyze what challenges may occur such as technical, social, and ethical barriers in the development and application of VR technology. Using a mixed methods approach, we collected quantitative data from Kaggle's dataset and conducted in-depth interviews with active VR users. Our findings show that although VR has significant potential in various fields, such as education, healthcare, and entertainment, challenges such as motion sickness, high costs, and limited and engaging content hinder the widespread adoption of VR technology. The study also highlights the importance of improving technologies that need to be implemented such as visual capabilities, motion responsiveness, and audio quality to enhance the user experience when using VR. By overcoming these barriers, VR can maximize its positive impact and become a revolutionary tool in various sectors. © 2024 The Authors.

关键词： Virtual environments

来源：评论

学校读者我要写书评

暂无评论

A fine-tuned vision transformer-based on limited dataset for facial expression recognition 9th

A fine-tuned vision transformer-based on limited dataset for...

引用

9th International Conference on computer Science and Computational Intelligence, ICCSCI 2024

作者： Febrian, Rio Huang, Ronald Richie Setiono, Nicholas Ramdhan, Dimas Chowanda, Andry Computer Science Department School of Computer Science Bina Nusantara University Jakarta11480 Indonesia Game Application & Technology Program Computer Science Department School of Computer Science Bina Nusantara University Jakarta11480 Indonesia

With the increase of interest in Facial Expression Recognition (FER) in the past few decades. Several challenges surfaced with the invention of many different FER models which are often based on Convolution Neural Network (CNN) architectures. Recently, an attention-based transformer model has been presented to address FER. One of the major issues with Transformers is the need for a large data quantity for training. Therefore, in this paper, we propose to learn how to fine-tune a vision transformer-based (ViT) model using a limited dataset. We will be using the JAFFE Dataset, which consists of only 213 images containing seven different emotions. The proposed method is evaluated using several fine-tuning methods, such as adding dropout, data augmentation, and layer freezing. We compared the models implemented with 5% dropout regularization, augmented dataset (up to 5000 images), and freezing the initial model's layers, fine-tuning around a fourth of the last layers. The best model was achieved by fine-tuning ViT L-16 with 96.06% accuracy, trained with 5% dropout in the augmented dataset, and freezing the initial 21st layers. We also compared our model to the other previous work model and the results showed that our model reached the state-of-the art for the JAFFE dataset. © 2024 The Authors.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Quantitative analysis of sign language translation using artificial neural network model 9th

Quantitative analysis of sign language translation using art...

引用

9th International Conference on computer Science and Computational Intelligence, ICCSCI 2024

作者： Wijaya, Fendy Dahendra, Leonardo Purwanto, Eko Setyo Ario, Muhamad Keenan Computer Science Department School of Computer Science Bina Nusantara University Jakarta11480 Indonesia Mobile Application & Technology Program Computer Science Department School of Computer Science Bina Nusantara University Jakarta11480 Indonesia

Sign language is one of the technique to support communication with deaf and speech impaired people. Nowadays, human needs become more complex, so are the needs of people with those disabilities. Therefore, with the sophistication of modern technology in the field of computer science, it is necessary to have a sign language translation system which is capable to convert human gestures into words or spelling letters in a natural human language, especially Indonesian language and SIBI sign language. This research is conducted to analyze the capabilities of an Artificial Neural Network (ANN) system to translate SIBI sign language into Indonesian language. The analysis is performed quantitatively from the experiment results to gain a descriptive insights for the generated models. By creating two datasets containing alphabets and words in SIBI sign language, two models were generated to predict alphabets and words, respectively. The result shows that ANN model could effectively and efficiently perform a sign language translation with accuracy of alphabet prediction reached 96.15% and words prediction reached 99.45% in duration of prediction without exceeding 0.15 seconds. Regarding to the results, it can be concluded that ANN model quantitatively suitable to be implemented as an Indonesian sign language translator system. © 2024 The Authors.

关键词： Neural machine translation

来源：评论

学校读者我要写书评

暂无评论

Multimodal emotion recognition model via hybrid model with improved feature level fusion on facial and EEG feature set

引用

Multimedia Tools and applications 2025年第1期84卷 1-36页

作者： Singh, Pratima Tripathi, Mukesh Kumar Patil, Mithun B. Shivendra Neelakantappa, Madugundu Department of Computer Science and Engineering Galgotias University Gautam Buddh Nagar Uttar Pradesh India Department of Computer Science and Engineering Vardhaman College of Engineering Telangana Hyderabad India Department of Computer Science and Engineering N K Orchid College of Engineering & Technology Maharastra Solapur India Department of Computer Application D. K. College Dumraon Bihar Buxar India Department of Information Technology Vasavi College of Engineering Telangana Hyderabad India

In recent years, academics have placed a high value on multi-modal emotion identification, as well as extensive research has been conducted in the areas of video, text, voice, and physical signal emotion detection. This paper proposes a novel multimodal emotion recognition model that employs a hybrid model with AMIG-based feature fusion on facial and EEG feature sets. The EEG signal and facial image are subjected to preprocessing to remove unwanted background noises with the Butterworth filter and Viola Jones Algorithm, respectively. While considering the pre-processed EEG signal, features such as EWFS-transform, wavelet features, and CSP-based features are extracted. In the proposed EWFS-transform, the window function is modified by using the frequency function and updated STFT. Conversely, the features including SE-AMM-EST-based features;LGXP and GLCM are extracted while considering the preprocessed face image. In the proposed SE-AMM-EST-based features, the mean of shape is updated and covariance in PCA for extracting texture-based features. In order to extract redundant free essential features, AMIG-based feature fusion is proposed. Further, the fused features are fed into a proposed hybrid model that includes LSTM and AML-CNN models. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.

关键词： Textures

来源：评论

学校读者我要写书评

暂无评论

Through the Lens: Unveiling the Power and Promise of Google's Visual Search technology 9

Through the Lens: Unveiling the Power and Promise of Google'...

引用

9th International Conference on Information Management and technology, ICIMTech 2024

作者： Salim, Devano Hofker Susanto, Jose Juan Manalu, Sonya R. Shiddiqi, Hafizh Ash School of Computer Science Bina Nusantara University Computer Science Department Jakarta11480 Indonesia School of Computer Science Bina Nusantara University Mobile Application & Technology Program Computer Science Department Jakarta11480 Indonesia

ISBN: (纸本)9798350390025

The digital landscape is undergoing profound changes, with visual search technology like Google Lens emerging as a pivotal force shaping our interaction with information. This research paper delves into Google Lens, a cutting-edge tool redefining how we engage with visual data. We comprehensively examine its functionalities, underlying principles, and real-world applications, particularly in education and library services. Through meticulous analysis, we evaluate its efficacy in addressing practical challenges and assess its accuracy, efficiency, and breadth of capabilities compared to other visual search technologies. By exploring existing research and the intricacies of Google Lens, this paper aims to provide a holistic understanding of this technology and its potential impact on how we access and interact with information in the real world. Finally, we explore the future of technology by contemplating the prospective roles of Google Lens across various industries and its potential for shaping the future of information access. © 2024 IEEE.

关键词： future of technology google lens visual search technology

来源：评论

学校读者我要写书评

暂无评论

Harnessing Deep Learning for Ocular Disease Diagnosis 9th

Harnessing Deep Learning for Ocular Disease Diagnosis

引用

9th International Conference on computer Science and Computational Intelligence, ICCSCI 2024

作者： Ryan, Jessica Nathaniel, Dave Andrew Purwanto, Eko Setyo Ario, Muhamad Keenan Computer Science Department School of Computer Science Bina Nusantara University Jakarta11480 Indonesia Mobile Application & Technology Program Computer Science Department School of Computer Science Bina Nusantara University Jakarta11480 Indonesia

Vision impairment, often caused by preventable ocular diseases can be challenging to diagnose accurately and prone to human error. Automation using technology, particularly deep learning, offers a promising solution to aid in accurate and efficient disease detection. This study explores the use of different CNN models specifically VGG-16, VGG-19, ResNet-50, and ResNet-152v2, for detecting ocular diseases. Simple fine-tuning is applied to these models, and their performance is compared to identify the most effective model. The purpose is to show how different models contribute to establishing reliable illness detection systems. The results reveal that most of these models perform well with even minimal fine-tuning. Among the models, ResNet-152v2 achieved the highest training accuracy of 90.36% demonstrating its capacity to learn from the training data. In contrast, ResNet-50 offered a more balanced performance with marginally lower accuracy, making it a robust choice for general application. © 2024 The Authors.

关键词： classification deep learning ocular diseases

来源：评论

学校读者我要写书评

暂无评论

Real Time Multiple-Object Detection Based on Enhanced SSD 2

Real Time Multiple-Object Detection Based on Enhanced SSD

引用

2nd International Conference on Power, Control and Computing Technologies, ICPC2T 2022

作者： Jaiswal, Tarun Pandey, Manju Tripathi, Priyanka National Institute of Technology Department of Computer Application Raipur India

ISBN: (纸本)9781665458580

Object detection and recognition are widely employed in a variety of computer vision applications, and numerous techniques have been suggested in the literature. Object detection is one of the areas of computer vision that has seen a lot of achievement. In the last decade, various applications of computer vision have been presented. Some of these include autonomous cars, video surveillance, and scene interpretation. Visual recognition systems, including image categorization, localization, and detection, are at the heart of all of these applications and have gathered a lot of research attention. These visual identification algorithms have achieved extraordinary performance because of considerable advancements in neural networks, particularly deep learning (DL). Object detection techniques in real-time systems, on the other hand, necessitate minimal computing time and a thorough performance assessment prior to deployment. In this paper multiple image detection approach is used to identify and locate the object concerning their class label, and an enhanced feature map is used for object detection. All this is possible due to the recent advancement in DL along with image processing. These images, fed to the model, come from the video source. To tackle the problem related to the aspect ratio, the model uses the multi-scale feature map together with a distinct filter for diverse default boxes. © 2022 IEEE.

关键词： computer vision Visualization System performance Power control Object detection Streaming media Feature extraction

来源：评论

学校读者我要写书评

暂无评论

Multi-scale Feature Fusion for Enhanced Person Re-identification 3rd

Multi-scale Feature Fusion for Enhanced Person Re-identific...

引用

3rd International Conference on Machine Learning, Cloud Computing and Intelligent Mining, MLCCIM 2024

作者： Wang, XiaoHui Sun, Yulin Zou, Zhipeng Liang, Xiaoyang Sun, Zhenyu Liu, Wei Shandong Business Institute Yantai China Shandong Xuewen Information Technology Co. Ltd. Jining China Technology of Computer Application Yantai Engineering and Technology College Yantai China School of Computer and Artificial Intelligence Zhengzhou University Zhengzhou China

ISBN: (纸本)9789819624676

In the realm of surveillance and security, person re-identification (Re-ID) stands as a critical challenge, particularly across non-overlapping camera networks. The essence of this challenge lies in the accurate and efficient identification of individuals by leveraging features captured from multiple perspectives. Addressing this, we introduce the Contextual Multi-Scale Network (Asy-CMSNet), a novel framework designed to enhance the Re-ID process through the integration of asymmetrical convolution and an innovative loss function, the Enhanced TriHard (E-TriHard). Unlike traditional approaches that rely on the stacking of convolutional layers, Asy-CMSNet employs a sophisticated mechanism that allows for the nuanced extraction and integration of multi-scale features, thereby optimizing the identification process across diverse pedestrian appearances. Central to our framework is the notion that effective person Re-ID necessitates the consideration of both global and localized pedestrian details, ranging from broad clothing characteristics to subtle accessories, which our model adeptly captures and utilizes. Through extensive experimentation across several benchmark datasets, Asy-CMSNet demonstrates superior performance in person Re-ID challenges, highlighting a substantial progression in the area. Our findings not only set new performance benchmarks but also pave the way for future research into more efficient, scalable, and robust person Re-ID systems. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Benchmarking

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：