Article
Computer Science, Artificial Intelligence
Vittorio Mazzia, Simone Angarano, Francesco Salvetti, Federico Angelini, Marcello Chiaberge
Summary: This research introduces an attention-based Action Transformer (AcT) architecture that outperforms current mix networks in human action recognition, leveraging small temporal windows of 2D pose representations for low-latency real-time performance. Additionally, a large-scale dataset called MPOSE2021 has been open-sourced to serve as a benchmark for training and evaluating real-time, short-time HAR. The proposed methodology was extensively tested on MPOSE2021, showcasing the effectiveness of the AcT model and setting the groundwork for future HAR research.
PATTERN RECOGNITION
(2022)
Article
Mathematics
Jinyoon Park, Chulwoong Kim, Seung-Chan Kim
Summary: Previous research on 3D skeleton-based human action recognition has often relied on sequence-wise viewpoint normalization process to adjust view directions. However, this approach has limitations in capturing the intricacies of complex action sequences. To address this, a straightforward sequence-wise augmentation technique is proposed to enhance the robustness of action recognition models.
Article
Computer Science, Information Systems
Faheem Shehzad, Muhammad Attique Khan, Muhammad Asfand E. Yar, Muhammad Sharif, Majed Alhaisoni, Usman Tariq, Arnab Majumdar, Orawit Thinnukool
Summary: Human action recognition based on Artificial intelligence reasoning is the most important research area in computer vision. This work proposes a deep learning and improved whale optimization algorithm based framework for HAR, which includes stages like pre-processing, transfer learning, feature fusion, and feature selection using modified serial approach and improved whale optimization algorithm. The proposed method achieves high testing accuracy on four datasets and outperforms state-of-the-art techniques.
CMC-COMPUTERS MATERIALS & CONTINUA
(2023)
Article
Computer Science, Information Systems
Sampat Kumar Ghosh, M. Rashmi, Biju R. Mohan, Ram Mohana Reddy Guddeti
Summary: This paper proposes a multi-stream Convolutional Neural Network model for multi-view human action recognition. The model utilizes depth and skeleton data, and introduces a novel and efficient depth descriptor and skeleton descriptor. Experimental results demonstrate that the proposed system outperforms state-of-the-art methods.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Review
Computer Science, Artificial Intelligence
Yuling Xing, Jia Zhu
Summary: Action recognition based on 3D skeleton data is a widely studied topic in computer vision, with the advantages of combining skeleton data and deep learning being gradually demonstrated. Previous research has mainly focused on video or RGB data methods, while GCN-based methods are gaining attention.
CAAI TRANSACTIONS ON INTELLIGENCE TECHNOLOGY
(2021)
Review
Computer Science, Hardware & Architecture
Xiankai Huang, Zhibin Cai
Summary: This paper summarizes and analyzes existing video action recognition methods based on 3D convolution to help new researchers understand this field. Firstly, it introduces the classical methods and points out the problems. Then, it summarizes the existing improved methods and compares the experimental results on benchmarks, discussing current challenges and future trends.
COMPUTERS & ELECTRICAL ENGINEERING
(2023)
Article
Computer Science, Interdisciplinary Applications
Duarte Moutinho, Luis F. Rocha, Carlos M. Costa, Luis F. Teixeira, Germano Veiga
Summary: This paper proposes a cognitive system powered by computer vision and deep learning to interpret implicit communication cues of the operator, aiming to increase the natural collaboration level of a robotic engine assembly station.
ROBOTICS AND COMPUTER-INTEGRATED MANUFACTURING
(2023)
Article
Computer Science, Artificial Intelligence
Onur Can Kurban, Nurullah Calik, Tulay Yildirim
Summary: This paper proposes a new temporal template approach for action recognition and person identification, which utilizes motion sequence information from masked depth video streams. The method creates a membership function to model motion changes and generates energy images to highlight intervals of motion with more change.
PATTERN RECOGNITION
(2022)
Article
Computer Science, Information Systems
Hussein Samma, Ali Salem Bin Sama
Summary: This research introduces a lightweight vision system with an optimized SqueezeNet backbone feature extraction network by integrating a two-layer particle swarm optimizer (TLPSO) into YOLO. It achieves this without sacrificing accuracy as the high-dimensional SqueezeNet convolutional filter selection is supported by the efficient TLPSO algorithm. The suggested system has shown a sevenfold boost in detection speed in recognizing human behaviors from drone-mounted camera images and outperformed earlier vision systems.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Wei Peng, Jingang Shi, Tuomas Varanka, Guoying Zhao
Summary: This article discusses the task of action recognition based on skeleton data and the mainstream framework ST-GCN, proposing a simple and effective strategy in experiments to capture global graph correlations, reducing model complexity, and achieving superior performance.
Article
Computer Science, Artificial Intelligence
Li Zhang, Chee Peng Lim, Yonghong Yu
Summary: The research proposes an ensemble model of evolving deep networks comprising Convolutional Neural Networks (CNNs) and bidirectional Long Short-Term Memory (BLSTM) networks using a swarm intelligence (SI) algorithm to determine optimal hyper-parameters for accurate representation of temporal dynamics of human actions. The SI algorithm incorporates hybrid crossover operators and a versatile search process to overcome stagnation, showing superiority over other methods for solving high-dimensional optimization functions.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Engineering, Electrical & Electronic
Haoran Wang, Baosheng Yu, Jiaqi Li, Linlin Zhang, Dongyue Chen
Summary: Skeleton-based human action recognition has received extensive attention due to its efficiency and robustness. However, it fails to recognize human actions induced by the interaction between human and objects. This paper introduces the multi-stream interaction networks (MSIN) to explore the dynamics of human skeleton, objects, and the interaction between human and objects simultaneously.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Article
Chemistry, Analytical
Seemab Khan, Muhammad Attique Khan, Majed Alhaisoni, Usman Tariq, Hwan-Seung Yong, Ammar Armghan, Fayadh Alenezi
Summary: Human action recognition (HAR) is crucial for smart surveillance systems but poses challenges due to the variety of actions and large video sequences. Deep learning (DL) systems have shown significant success in HAR, achieving high accuracies on multiple datasets. The proposed DL-based design includes feature mapping, fusion, selection steps, and outperforms state-of-the-art methods in terms of computational time.
Review
Chemistry, Analytical
Md Golam Morshed, Tangina Sultana, Aftab Alam, Young-Koo Lee
Summary: Human action recognition systems use sensor data to accurately identify and interpret human actions. Recent developments in computer vision, particularly deep learning-based features, have significantly improved action recognition. This study provides in-depth analysis of human activity recognition, including the latest research on acquiring human action features and deep learning techniques.
Article
Automation & Control Systems
Dawei Zhang, Yanmin Zhang, Meng Zhou
Summary: This article presents a skeleton-guided action recognition framework with multistream 3D convolutional neural network for elderly-care robot. It enhances the feature extraction ability of human action and improves recognition accuracy through the use of parallel dual-stream networks and decision fusion. The experimental results demonstrate its superior performance.
ADVANCED INTELLIGENT SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Marcin Wozniak, Jakub Silka, Michal Wieczorek
Summary: Modern medical clinics use computer systems to support medical examinations and detect potential health problems more efficiently. Deep learning approaches have been proven to provide the most precise results in evaluating CT brain scans. In this article, a novel correlation learning mechanism (CLM) is proposed to combine convolutional neural network (CNN) with classic architecture. The support neural network helps CNN to optimize pooling and convolution layers, resulting in faster learning and higher efficiency of the main neural classifier.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Ashis Paul, Arpan Basu, Mufti Mahmud, M. Shamim Kaiser, Ram Sarkar
Summary: This article discusses the use of deep learning models and an inverted bell-curve weighted ensemble method to assist in the detection of COVID-19 in CXR images. By using transfer learning and retraining models pretrained on the ImageNet dataset, as well as performing weighted average predictions, the accuracy of COVID-19 identification in CXR images can be improved.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Computer Science, Software Engineering
Samir Malakar, Samanway Sahoo, Anuran Chakraborty, Ram Sarkar, Mita Nasipuri
Summary: Handwritten word recognition is an open research problem due to variations in writing style and degraded images. This paper proposes a holistic approach combined with distance calculation and feature descriptors to address the problem. The experimental results demonstrate the effectiveness of the proposed method on standard databases compared to deep learning models.
Article
Computer Science, Artificial Intelligence
Souradeep Mukhopadhyay, Sabbir Hossain, Samir Malakar, Erik Cuevas, Ram Sarkar
Summary: This paper introduces a new gray-scale contrast enhancement algorithm, which improves image quality by calculating near-optimal values using the Artificial Electric Field Algorithm (AEFA). Through comparisons with other techniques using standard metrics, simulation results show that the proposed method increases image contrast and enriches image information.
Article
Computer Science, Information Systems
Anubhab Das, Arka Choudhuri, Arpan Basu, Ram Sarkar
Summary: This study proposes a GAN-based method for generating handwritten Bengali compound characters to address data scarcity. The model's performance is evaluated by assessing the quality of generated samples, showing that it outperforms basic AC-GAN architecture and some other existing GAN architectures.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Correction
Computer Science, Artificial Intelligence
Apu Sarkar, S. K. Sabbir Hossain, Ram Sarkar
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Sk Mohiuddin, Khalid Hassan Sheikh, Samir Malakar, Juan D. Velasquez, Ram Sarkar
Summary: Digital face manipulation has become a significant concern recently due to its harmful effects on society, particularly for high-profile celebrities who can easily be targeted using apps like FaceSwap and FaceApp. Detecting deepfake images or videos is challenging, and existing models often fail to check for irrelevant or redundant features. In this study, a hierarchical feature selection (HFS) method using a hybrid population-based meta-heuristic model and a single solution-based meta-heuristic model was proposed. The model achieved high AUC scores on three publicly available datasets and outperformed most state-of-the-art methods.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Computer Science, Information Systems
Wei Wei, Qiao Ke, Dawid Polap, Marcin Wozniak
Summary: Digital security in modern systems often relies on biometric methods, and new implementations continue to emerge. This can be seen in various applications, such as signing for a courier package pick-up. However, signature verification is a complex process due to variations in size, angle, and writing conditions. Therefore, new methods are constantly needed to evaluate signatures. In this article, the authors propose the use of spline interpolation and two types of artificial neural networks to verify the identity of a person based on selected local and global features extracted from signature images. Experimental results on the SVC2004 database demonstrate an accuracy of 87.7%.
IEEE INTERNET OF THINGS JOURNAL
(2023)
Article
Medicine, General & Internal
Arnab Bagchi, Payel Pramanik, Ram Sarkar
Summary: Breast cancer is a deadly disease that affects women worldwide. Early diagnosis and proper treatment can save lives. Breast image analysis, including histopathological image analysis, and computer-aided diagnosis, can help improve efficiency and accuracy in breast cancer detection. In this study, a deep learning-based method was developed to classify breast cancer using histopathological images, achieving high classification accuracy.
Article
Computer Science, Artificial Intelligence
Qingsen Yan, Axi Niu, Chaoqun Wang, Wei Dong, Marcin Wozniak, Yanning Zhang
Summary: Deep learning-based methods have achieved remarkable results in the field of super-resolution. However, the limitation of paired training image sets has led researchers to explore self-supervised learning. However, the assumption of inaccurate downscaling kernel functions often leads to degraded results. To address this issue, this paper introduces KGSR, a kernel-guided network that trains both upscaling and downscaling networks to generate high-quality high-resolution images even without knowing the actual downscaling process.
PATTERN RECOGNITION
(2024)
Proceedings Paper
Computer Science, Artificial Intelligence
Alan Popiel, Marcin Wozniak
Summary: This paper presents a model and an algorithm to optimize the placement of routers in a network system for the mining industry, with N chambers and N-1 or fewer connections between them. The model considers two types of routers with different signal strengths, and the algorithm has a computational complexity of O(n(2), as tested on sample graph structures.
ARTIFICIAL INTELLIGENCE AND SOFT COMPUTING, ICAISC 2022, PT II
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Jakub Silka, Michal Wieczorek, Martyna Kobielnik, Marcin Wozniak
Summary: Deep learning architectures are used for demanding analysis of complex data inputs, where regular neural networks may encounter issues. In this article, we propose a deep learning model based on a BiLSTM neural network architecture. The proposed model is trained using the Adam algorithm, and we also examine other latest algorithms to determine the best configuration. Results show that our proposed BiLSTM deep learning neural network achieves over 99% accuracy.
ARTIFICIAL INTELLIGENCE AND SOFT COMPUTING, ICAISC 2022, PT I
(2023)
Article
Computer Science, Information Systems
Jyotismita Chaki, Marcin Wozniak
Summary: This study proposes a reinforcement learning agent that can interact with brain tumor images to retrieve and categorize similar images. The proposed method utilizes a novel architecture and binary coding technique, as well as fuzzy logic-based sample generation, to improve brain tumor classification and retrieval.
Article
Computer Science, Artificial Intelligence
Marcin Wozniak, Jozef Szczotka, Andrzej Sikora, Adam Zielonka
Summary: This article presents a model of adjustable moisture control for historical buildings, utilizing a flexible IoT infrastructure and type-2 fuzzy logic reasoning to create an innovative intelligent system for interior conditions control. The developed system, tested in an old brewery building, showed efficient dehumidification results at a low cost.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Article
Computer Science, Information Systems
G. S. Nijaguna, N. Dayananda Lal, Parameshachari Bidare Divakarachari, Rocio Perez de Prado, Marcin Wozniak, Raj Kumar Patra
Summary: This research combines the Internet of Medical Things and artificial intelligence to develop a method for monitoring and diagnosing cardiac arrhythmia. By extracting various features from electrocardiogram signals and using an Auto Encoder and Selective Opposition algorithm, a classification system is built. The classified results are interpreted using the Shapley additive explanations. The experimental results show that the proposed method achieves higher accuracy compared to other existing methods.