Solving Partially Observable 3D-Visual Tasks with Visual Radial Basis Function Network and Proximal Policy Optimization DOI Creative Commons

Julien Hautot,

Céline Teulière, Nourddine Azzaoui

et al.

Machine Learning and Knowledge Extraction, Journal Year: 2023, Volume and Issue: 5(4), P. 1888 - 1904

Published: Dec. 1, 2023

Visual Reinforcement Learning (RL) has been largely investigated in recent decades. Existing approaches are often composed of multiple networks requiring massive computational power to solve partially observable tasks from high-dimensional data such as images. Using State Representation (SRL) shown improve the performance visual RL by reducing into compact representation, but still relies on deep and environment. In contrast, we propose a lighter, more generic method extract sparse localized features raw images without training. We achieve this using Radial Basis Function Network (VRBFN), which offers significant practical advantages, including efficient accurate training with minimal complexity due its two linear layers. For real-world applications, scalability resilience noise essential, real sensors subject change noise. Unlike CNNs, may require extensive retraining, network might only need minor fine-tuning. test efficiency VRBFN representation different Proximal Policy Optimization (PPO). present large study comparison our extraction methods five classical SRL first-person scenarios. show that approach presents appealing sparsity robustness obtained results when agents better than other tested four proposed

Language: Английский

Validity of AI-Based Gait Analysis for Simultaneous Measurement of Bilateral Lower Limb Kinematics Using a Single Video Camera DOI Creative Commons
Takumi Ino, Mina Samukawa, Tomoya Ishida

et al.

Sensors, Journal Year: 2023, Volume and Issue: 23(24), P. 9799 - 9799

Published: Dec. 13, 2023

Accuracy validation of gait analysis using pose estimation with artificial intelligence (AI) remains inadequate, particularly in objective assessments absolute error and similarity waveform patterns. This study aimed to clarify measures for pattern AI (OpenPose). Additionally, we investigated the feasibility simultaneous measuring both lower limbs a single camera from one side. We compared motion data video footage that was synchronized three-dimensional device. The comparisons involved mean (MAE) coefficient multiple correlation (CMC) compare similarity. MAE ranged 2.3 3.1° on side 3.1 4.1° opposite side, slightly higher accuracy Moreover, CMC 0.936 0.994 0.890 0.988 indicating "very good excellent" Gait revealed precision sides sufficiently robust clinical evaluation, while measurement superior

Language: Английский

Citations

19
Violence-YOLO: Enhanced GELAN Algorithm for Violence Detection DOI Creative Commons
Wenbin Xu,

Dingju Zhu,

Renfeng Deng

et al.

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(15), P. 6712 - 6712

Published: Aug. 1, 2024

Violence is a serious threat to societal health; preventing violence in airports, airplanes, and spacecraft crucial. This study proposes the Violence-YOLO model detect accurately real time complex environments, enhancing public safety. The based on YOLOv9’s Generalized Efficient Layer Aggregation Network (GELAN-C). A multilayer SimAM incorporated into GELAN’s neck identify attention regions scene. YOLOv9 modules are combined with RepGhostNet GhostNet. Two modules, RepNCSPELAN4_GB RepNCSPELAN4_RGB, innovatively proposed introduced. shallow convolution backbone replaced GhostConv, reducing computational complexity. Additionally, an ultra-lightweight upsampler, Dysample, introduced enhance performance reduce overhead. Finally, Focaler-IoU addresses neglect of simple difficult samples, improving training accuracy. datasets derived from RWF-2000 Hockey. Experimental results show that outperforms GELAN-C. [email protected] increases by 0.9%, load decreases 12.3%, size reduced 12.4%, which significant for embedded hardware such as Raspberry Pi. can be deployed monitor places effectively handling backgrounds ensuring accurate fast detection violent behavior. In addition, we achieved 84.4% mAP Pascal VOC dataset, reduction parameters compared previously refined detector. offers insights real-time behaviors environments.

Language: Английский

Citations

3
Designing and Manufacturing an Affordable and Easy to Use Visual Bio Feedback Device to Fix Forward Head Posture: A Pilot Study Involving Female Students DOI Creative Commons
Mehran Emadi Andani,

Bahar Lotfalian,

Ata Jahangir Moshayedi

et al.

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(2), P. 781 - 781

Published: Jan. 17, 2024

Forward Head Posture (FHP) is when the head leans forward due to factors such as heavy backpacks or poor computer ergonomics. FHP can lead neck strain and discomfort well potential long-term issues arthritis. Treatment options include specialized exercises, orthopedic devices, manual therapy, physical visual feedback techniques, along with guidance from specialists in medicine rehabilitation. In this study, a feedback-based approach was used address female students. The study spanned ten days included group control group. results showed significant improvements maximum angle deviation compared group; however, there no change DFA number, indicating stability policy by central nervous system. demonstrated that sessions led immediate benefits, participants progressively acquiring skills involving maintenance of proper positioning. test indicated decreased less than 15 degrees, return normal state. versatility developed affordable easy-to-use device for using smartphone motion sensors similar systems are discussed paper well. suggests promising healthcare, including remote monitoring smartphone-based solutions.

Language: Английский

Citations

2
Empirical Analysis of Widely Used Website Automated Testing Tools DOI Creative Commons

Balqees Sani,

Sadaqat Jan

EAI Endorsed Transactions on AI and Robotics, Journal Year: 2024, Volume and Issue: 3

Published: Oct. 10, 2024

In today's software development, achieving product quality while minimising cost and time is critical. Automated testing crucial to attaining these goals by lowering inspection efforts discovering faults more effectively. This paper compares widely used automated tools, such as Selenium, Appium, Java Unit (JUnit), Test Next Generation (TestNG), Jenkins, Cucumber, LoadRunner, Katalon Studio, Simple Object Access Protocol User Interface (SoapUI), TestComplete, based on functionality, ease of use, platform compatibility, integration capabilities. Our findings show that no single tool inherently superior, with each excelling in certain areas online, mobile, Application Programming (API), or performance testing. While Selenium Appium are the dominant online mobile frameworks, TestComplete Studio offer complete, user-friendly cross-platform solutions. Despite benefits automation, obstacles maintenance, scalability, issues remain. The report finishes advice for picking best project offers potential approaches enhancing AI-driven optimisation, cloud-based testing, greater Continuous Integration/ Deployment (CI/CD) integration. study useful information developers testers looking optimise their methods increase quality.

Language: Английский

Citations

2
Underwater smart glasses: A visual-tactile fusion hazard detection system DOI Creative Commons

Zhongze Ma,

Chenjie Zhang, Pengcheng Jiao

et al.

iScience, Journal Year: 2024, Volume and Issue: 27(4), P. 109479 - 109479

Published: March 11, 2024

Marine activities typically face various risk factors such as marine animal attacks or unexpected collisions. In this paper, we develop underwater smart glasses (USGs) based on visual-tactile fusion for hazard detection in real-time, ensuring operational safety. The proposed USG is composed of the vision module by artificial intelligence (AI)-enabled optical sensing and tactile triboelectric metamaterials-enabled mechanical sensing. obtained target algorithm

Language: Английский

Citations

1
Repmono: a lightweight self-supervised monocular depth estimation architecture for high-speed inference DOI Creative Commons
Guowei Zhang,

Xincheng Tang,

Li Wang

et al.

Complex & Intelligent Systems, Journal Year: 2024, Volume and Issue: 10(6), P. 7927 - 7941

Published: Aug. 10, 2024

Self-supervised monocular depth estimation has always attracted attention because it does not require ground truth data. Designing a lightweight architecture capable of fast inference is crucial for deployment on mobile devices. The current network effectively integrates Convolutional Neural Networks (CNN) with Transformers, achieving significant improvements in accuracy. However, this advantage comes at the cost an increase model size and reduction speed. In study, we propose named Repmono, which includes LCKT module large convolutional kernel RepTM based structural reparameterisation technique. With combination these two modules, our achieves both local global feature extraction smaller number parameters significantly enhances Our network, 2.31MB parameters, shows accuracy over Monodepth2 experiments KITTI dataset. uniform input dimensions, network's speed 53.7% faster than R-MSFM6, 60.1% Monodepth2, 81.1% MonoVIT-small. code available https://github.com/txc320382/Repmono .

Language: Английский

Citations

1
Multi-Adjacent Camera-Based Dangerous Driving Trajectory Recognition for Ultra-Long Highways DOI Creative Commons
Liguo Zhao,

Zhipeng Fu,

Jingwen Yang

et al.

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(11), P. 4593 - 4593

Published: May 27, 2024

Fast detection of the trajectory is key point to improve further emergency proposal. Especially for ultra-long highway, prompt labor-intensive. However, automatic relies on accuracy and speed vehicle detection, tracking. In multi-camera surveillance system highways, it often difficult capture same without intervals, which makes re-recognition crucial as well. this paper, we present a framework that includes tracking using improved DeepSORT, re-identification, feature extraction based rules, behavior recognition analysis. particular, design network architecture DeepSORT with YOLOv5s address need real-time in real-world traffic management. We an attribute module generate matching individuality attributes vehicles re-identification performance under multiple neighboring cameras. Besides, use bidirectional LSTM improves prediction, demonstrating its robustness noise fluctuations. The proposed model has high advantage from cumulative characteristic (CMC) curve shown even above 15.38% compared other state-of-the-art methods. developed local highway dataset comprehensively evaluated, including abnormal recognition, lane change anomaly recognition. Experimental results demonstrate effectiveness method accurately identifying various behaviors, changes, stops, dangerous driving behavior.

Language: Английский

Citations

0
Impact of Regular Sleep Patterns on Learning Achievement in Preschool Children in Mindahan Batealit: Issues and Implications DOI Open Access

Helmi Sintha Faradila Purnama,

Muhammad Nofan Zulfahmi

AS-SABIQUN, Journal Year: 2024, Volume and Issue: 6(4), P. 819 - 830

Published: July 1, 2024

Regular sleep patterns are one of the crucial factors that can affect children's development, including in terms learning achievement. At preschool age, children at a very important stage where they experience significant physical, cognitive, and emotional growth. The study aimed to examine preschool-aged Dukuh Mindahan Lor Batealit Jepara. Research was limited subjects were preschool-age Jepara object this This uses qualitative research methods obtain an in-depth description children, with case approach using information collection techniques form interviews, observations documentation analyzing describing village. results show positive correlation between optimal duration (9-10 hours per night) achievement preschoolers. Children who get enough tend have higher cognitive scores better concentration skills class.

Language: Английский

Citations

0
Mapping Generative Artificial Intelligence (GAI's) Exciting Future: From Gemini to Q* and Beyond DOI Creative Commons
Zarif Bin Akhtar

EAI Endorsed Transactions on AI and Robotics, Journal Year: 2024, Volume and Issue: 3

Published: Aug. 15, 2024

This research investigates the transformative potential of Mixture Experts (MoE) and multimodal learning within generative AI, exploring their roles in advancing towards Artificial General Intelligence (AGI). By leveraging a combination specialized models, MoE addresses scalability computational limitations, enabling more nuanced robust modelling across diverse data modalities. The exploration draws inspiration from pioneering projects like Google's Gemini OpenAI's anticipated Q* to push boundaries AI capabilities. objectives include impact on investigating learning's role achieving AGI, conducting experiments demonstrate MoE's effectiveness various domains, assessing influence AI-generated preprints peer-review process. Ethical considerations are also emphasized, advocating for development that aligns with societal well-being. methodology employs techniques social network analysis examine current landscape future possibilities learning. Experiments conducted healthcare, finance, education 25% increase training efficiency 30% improvement output quality when using compared traditional single-model approaches. highlights significant process scholarly communication. findings underscore propel AGI. study advocates responsible development, aligned human-centric values well-being, proposes strategic directions research. promotes balanced ethical integration MoE, multimodality, AGI fostering equitable distribution usage technologies.

Language: Английский

Citations

0
Point Cloud Clustering Segmentation Algorithm for Indoor and Outdoor Environments DOI
Shaohu Wang, Aiguo Song, Tianyuan Miao

et al.

Published: June 27, 2024

Language: Английский

Citations

0