Earticle

The researcher set up multiple sensor units on the road slope such as national highway and highway where there is a possibility of loss, and using the acceleration sensor built into the sensor unit the researcher will sense whether the inclination of the road slope occur in real time, and Based on the sensed data, the researcher tries to implement a system that detects collapse of road slope and dangerous situation. In the experiment of measuring the error between the actual measurement time and the judgment time of the monitoring system when judging the warning of the sensor and falling rock detection by using the acceleration sensor, the error between measurement time and the judgment time at the sensor warning was 0.34 seconds on average, and an error between measurement time and judgment time at falling rock detection was 0.21 seconds on average. The error is relatively small, the accuracy is high, and thus the change of the slope can be clearly judged.

얼굴 검출 및 정렬은 많은 어플리케이션에서 사용되며 딥러닝의 발전으로 성능이 많이 향상되었다. 하지 만 대부분의 최신 얼굴 검출 및 정렬 방법들은 매우 많은 연산을 필요로 하여 자원이 제한되는 모바일 환경이나 임베디드 시스템과 같은 엣지 장치에서 활용이 어려운 문제가 있다. 이를 해결하기 위해 본 논문에서는 EdgeFace 라는 새로운 얼굴 검출 및 정렬 기법을 제안하여 엣지 장치에서 속도와 정확성 측면에서 우수한 성능을 달성한다. 0.3M 파라미터를 초과하지 않으며 기존 경량 백본 보다 우수한 정확도를 달성하는 EdgeFaceNet이라는 새로운 경량 백본을 설계하고 검출 네트워크 부분에서는 특징 맵 스케일링 및 앵커 박스 설정을 사용하여 정확도 손실 없이 속도를 개선하도록 최적화하였다. WIDER FACE Hard 데이터셋에 대한 실험에서 EdgeFace는 83.8%의 정 확도로 Nvidia Tesla K80 GPU 에서 120 FPS, Nvidia Jetson Nano에서 26 FPS의 속도를 달성하였다.

Although recent face detection tasks have made significant progress, applying them to edge devices is still a difficult task. In this paper, we propose a new face detector called EdgeFace to achieve excellent performance in terms of speed and accuracy on edge devices. We design a new lightweight backbone called EdgeFaceNet and present a channel attention-based detection network. The backbone does not exceed 0.3M parameters and achieves better accuracy than existing lightweight backbones. In the detection network part, feature map scaling and anchor box settings are used to optimize to improve speed with meaningful accuracy, and improve the utilization contextual by channel-attention. From experiments on WIDER FACE hard dataset, EdgeFace achieves state-of-the-art performance among the lightweight face detection methods and runs at 120 FPS on Nvidia Tesla K80 GPU and 26 FPS on Nvidia Jetson Nano.

감성기반 공간디자인 기술은 사용자가 감성에 영향을 미치는 요소들을 변화시킬 수 있고 사용자가 원하는 공간을 디자인 할 수 있다. 실제 공간을 구성하는데 있어서 제작 기간의 단축을 제공하고 제작된 콘텐츠의 품질도 보장할 수 있다. 감성기반의 디지털 공간디자인 시스템은 기본적으로 구축된 감성모형을 활용하여 사용자가 감성 공간을 디자인하는데 도움을 받거나 디자인 하고자 하는 콘텐츠의 품질을 평가하는데 활용 할 수 있다. 시스템에서 제공하는 평가시스템에 의한 실시간 평가가 가능하며 평가에 의해 축적되는 데이터를 효율적으로 관리할 수 있다. 평가 데이터를 인터넷의 중앙서버에 데이터베이스화하고 실시간 업데이트 데이터를 통해 객관적이고 정밀한 결과를 도출해 낼 수 있다. 감성기반 리얼타임 디지털 공간디자인시스템은 일반인도 쉽게 사용할 수 있는 인터페이스를 제공하여 가상현실, 게임, 애니메이션 등의 여러 분야의 공간디자인 개발에 효과적인 방법을 제공하고자 한다.

An emotion-based space design technology can change some factors with which a user may affect emotion, and enable a user to design a desired space. In composing an actual space, the production period may be reduced and the quality of produced contents may also be ensured. An emotion-based digital space design system can either help a user design an emotional space by utilizing a basically constituted emotion model or be utilized in evaluating the quality of contents for the design. Also, it is possible to make a real-time evaluation using an evaluation system offered by the system and to efficiently manage data accumulated by the evaluation. In addition, an objective and an accurate result are available to be drawn out through real-time updated data while databasing evaluation data on a central server on the Internet. The emotion-based real-time digital space design system aims to deliver effective methods for space design development in various fields, virtual reality, games, and animations by furnishing an interface a general user may easily use.

내장형 시스템에서 사용되는 많은 운영체제 중 윈도우즈는 실시간성 지원의 부재로 점검장비와 같은 실시간성이 필수적으로 요구되는 시스템에는 적합하지 않다. 이러한 결점을 보완하기 위한 기존의 서드파티들은(즉, RTX나 INTime) 고가의 구입비와 유지보수비로 인해 점검장비 프로그램 개발 시 비용의 증가를 초래하는 문제점이 있다. 본 논문에서는 윈도우즈가 사용하지 않는 멀티프로세서 기반 x86 아키텍처의 Local APIC를 이용하여 윈도우즈와는 독립적인 인터럽트를 발생시켜, 윈도우즈에 실시간성을 보장하는 RTiK을 설계 및 구현 하였으며, 또한 인터럽트 지연시간을 줄이기 위해 윈도우즈에서 제공하는 지연처리호출을 사용하였다. 마지막으로 윈도우즈에 실시간성을 보장하는 실시간 이식커널의 성능을 커널레벨과 유저레벨에서 측정함으로써 제안한 RTiK의 성능을 검증한다.

With lack of real-time support, Windows is not appropriate for test equipments which inevitably require real-time support. Consequently we have no choice to use expensive third-party solutions such as RTX or INtime. In this paper, we design and implement a real-time implanted kernel(RTiK) which support real-time on Windows by using the local APIC of multiprocessor-based x86 architectures. To decrease the interrupt latency, we also use the deferred procedure calls supported by Windows. Finally, we evaluate the performance of the proposed RTiK by measuring real-time capacities of RTiK both on User- and Kernel-levels.

실시간 제어는 관리 시스템의 실질적인 사용을 확인하기 위해 해결해야 하는 주요 과제였다. 이와 관련하여 편의성 과 효율성을 높이기 위해 처음으로 사물인터넷(IoT) 기반 저수조 시스템을 제안 및 개발하였다. 저수조의 상태가 불안정할 경우 사용자에게 알려 저수조를 효과적으로 제어할 수 있다. 제안된 시스템은 센서 데이터 측정 및 제어를 위한 내장형 H/W 장치, 웹 및 모바일 앱을 통한 관리 서버 구축을 위한 애플리케이션 S/W, 통계 관리 및 모니터링을 위한 효율적인 데이터베 이스 구조로 구성되어 있다. 또한 기계 학습 알고리즘을 적용하여 실제 효율성을 더욱 향상시킬 수 있다.

Real-time controllability has been a major challenge that should be addressed to ascertain the practical usage of the management systems. In this regards, for the first time, we proposed and implemented an IoT(Internet of Things)-based water tank system to improve convenience and efficiency. The reservoir can be effectively controlled by notifying the user if the condition of the reservoir is unstable. The proposed system consists of embedded H/W unit for sensor data measuring and controling, application S/W for deployment of management server via web and mobile app, and efficient database structure for managing and monitoring statistics. And machine learning algorithms can be applied for further improvements of efficiency in practice.

Safe operation of freight vehicles is an important issue for drivers, cargo, and other road users. In particular, the center of gravity of a freight vehicle is directly related to the stability of the vehicle, and this can fluctuate in real time depending on weight changes. Every time a freight vehicle loads or unloads cargo, its center of gravity changes, and these changes greatly affect the risk of vehicle rollover. We researched a continuous center of gravity measurement system for freight vehicles for safe driving.

Optical camera communication (OCC) leverages camera image sensors for data reception from light sources but faces challenges of low data rates and high bit error rates. This study introduces an OCC system combining orthogonal frequency division multiplexing with a UNet-based equalizer for signal denoising. Using pixel rows as transmission units, the system achieves a data rate of 9.2 kbps and a bit error rate of at 1 m. Python scripts facilitate system control, optimization, and embedded deployment, highlighting OCC’s potential for next-generation communication systems with improved performance over conventional methods.

The communication gap between the deaf and public is the concern for both parents and the government of Bhutan. The deaf school urges people to learn Bhutanese Sign Language (BSL) but learning Sign Language (SL) is difficult. This paper presents the BSL digits recognition system using the Convolutional Neural Network (CNN) and a first-ever BSL dataset which has 20,000 sign images of 10 static digits collected from different volunteers. Different SL models were evaluated and compared with the proposed CNN model. The proposed system has achieved 97.62% training accuracy. The system was also evaluated with precision, recall, and F1-score.

Nowadays, detection of license plate (LP) for non-helmeted motorcyclist has become mandatory to ensure the safety of the motorcyclists. This paper presents the real-time detection of LP for non-helmeted motorcyclist using the real-time object detector YOLO (You Only Look Once). In this proposed approach, a single convolutional neural network was deployed to automatically detect the LP of a non-helmeted motorcyclist from the video stream. The centroid tracking method with a horizontal reference line was used to eliminate the false positive generated by the helmeted motorcyclist as they leave the video frames. The overall LP detection rate was 98.52%.

The Automatic License Plate Recognition (ALPR) is one of the intelligent transportation systems which provides a safe and secure mode of transportation. In ALPR technology, recognition accuracy entirely depends on the performance of the localization phase. This paper presents the real-time Bhutanese license plate (LP) localization using YOLO (You Only Look Once). The vehicle detection was performed before the LP localization to eliminate the false positives generated by the signboards as they look similar to LPs. A single convolutional neural network gave an overall mean average precision of 98.6% with the training loss of 0.0231 for vehicle and LP.

Objective : Conventional therapy approaches for stroke survivors have required considerable demands on therapist’s effort and patient’s expense. Thus, new robotics rehabilitation therapy technologies have been proposed but they have suffered from less than optimal control algorithms. This article presents a novel technical healthcare solution for the real-time, simultaneous and propositional myoelectric control for stroke survivors’ upper limb robotic rehabilitation therapy. Methods : To implement an appropriate computational algorithm for controlling a portable rehabilitative robot, a linear regression model was employed, and a simple game experiment was conducted to identify its potential of clinical utilization. Results : The results suggest that the proposed device and computational algorithm can be used for stroke robot rehabilitation. Conclusion : Moreover, we believe that these techniques will be used as a prominent tool in making a device or finding new therapy approaches in robot-assisted rehabilitation for stroke survivors.

Objective To describe an ultrasonography-guided technique for cubital tunnel injection.Method The ulnar nerves from 12 elbows of 6 adult cadavers were scanned, and the cross-sectional areas of the ulnar nerves, cubital tunnel inlets and outlets were measured by using ultrasonography. All elbows were dissected after an ultrasonography-guided dye injection at the inlet of the cubital tunnel. The dissectors evaluated the spread of dye and the coloration of the nerve and remeasured the cross-sectional areas of the cubital tunnel inlets and outlets.Results After a real-time visualization of an ultrasonography-guided injection, the ulnar nerves were seperated from the medial groove for the ulnar nerve. All the ulnar nerves of the cadavers were successfully colored with the dye, from the inlet to oulet of the cubital tunnel. The post-injection cross-sectional areas were significantly larger than the pre-injection cross-sectional areas. No significant differences were detected in the post-injection cross-sectional areas of the cubital tunnel outlet and the ulnar nerve as compared with the pre-injection areas.Conclusion Clinicians should consider real-time visualization of ultrasonography for guided injection around the ulnar nerve at the inlet of the cubital tunnel.

Fall accidents are increasing, and monitoring them using real-time CCTV systems remains challenging. This paper compares the performance of YOLOv11 and RT-DETRv2 models for real-time fall detection. Experimental results show that YOLOv11 outperforms RT-DETRv2 in terms of inference speed, making it more suitable for real-time applications. Unlike earlier studies, we propose feature map-based knowledge distillation during the model training process to improve model performance. The proposed YOLO-based fall detection system transfers intermediate representations from a teacher to a student network and optimises two complementary objectives: spatial alignment via Mean-Squared-Error (MSE) loss and channel-wise distribution alignment via Kullback?Leibler (KL) divergence. Experiments improved the mean Average Precision (mAP) and reduced processing time by 0.8ms. Evaluation on AI-hub abnormal behavior datasets confirmed a 0.02 increase in accuracy and F1-score, demonstrating the effectiveness of the proposed distillation method in real-time environments.

The Maritime Tactical Data System is a software system that collects track data from maritime sensors to compile and show on a sea map to provide maritime patrol vessels or maritime surveillance stations with situational awareness. Speed and precision in tracking multiple targets are crucial in achieving situational awareness. A multi-target tracking problem is NP-hard if it involves more than two sensors and a large amount of data since it generates many potential solutions that must be evaluated. Previous research has demonstrated that the Density-based Spatial Clustering of Applications with Noise (DBSCAN) method can perform track-to-track association with pretty good results; nevertheless, the density-reachable concept of DBSCAN poses a problem when two targets are within a distance less than the threshold. Another limitation is the inability of DBSCAN to associate tracks as soon as sensor track data is received. DBSCAN must run after all data has been collected in a database. In this paper, a novel track-to-track association method called Neighborhood Clustering Track Association and Fusion (NCTAF) is proposed to address the limitations of DBSCAN. According to the experiment results, NCTAF overcame the inaccurate cluster form generated by DBSCAN. The most remarkable result is that NCTAF performs track associations in an average of one second after receiving sensor track data involving three sensors, 4000 track data per sensor, and an update rate of 5-12 s per sensor. In contrast, DBSCAN required more than 10 min for the same scenario.

Objective To investigate the feasibility of ultrasound (US)-guided steroid injection by in-plane approach for cubital tunnel syndrome (CuTS), based on symptomatic, morphologic and electrophysiological outcomes.Methods A total of 10 patients, who were clinically diagnosed as CuTS and confirmed by an electrodiagnostic study, participated in this study. US-guided injection into the cubital tunnel was performed with 40 mg triamcinolone and 2 mL of 1% lidocaine. Outcomes of the injections were evaluated at pre-injection, 1st week and 4th week after injection. Visual analog scale, self-administered questionnaire of the ulnar neuropathy at the elbow (SQUNE), and McGowan classification were used for clinical evaluation. Cross-sectional area of the ulnar nerve by US and the electrophysiological severity scale through a nerve conduction study were utilized in the evaluation of morphologic and electrophysiological changes. The cross-sectional area of the ulnar nerve was measured at 3 points of condylar, proximal, and distal level of the cubital tunnel.Results No side effects were reported during the study period. The visual analog scale and cross-sectional area showed a significant decrease at 1st week and 4th week, as compared to baseline (p<0.05). The electro-physiological severity scale was significantly decreased at the 4th week, as compared with baseline and 1st week (p<0.05). Among the quantitative components of the scale, there were statistically significant improvements with respect to the conduction velocity and block.Conclusion The new approach of US-guided injection may be a safe tool for the treatment of CuTS. Symptomatic and morphologic recoveries preceded the electrophysiological improvement.

As the number of low Earth orbit satellites is increasing, new algorithms and high-performance computing resources are needed to handle the enormous real-time arrival data generated by satellites. Based on the requirements, this paper proposes a novel real-time adaptive speckle filtering selection algorithm for satellite synthetic aperture radar (SAR) images. The use of high-performance filters generates high-quality images whereas it introduces delays. Thus, a real-time adaptive algorithm which achieves time-average SAR image quality maximization subject to delays using Lyapunov optimization, where the delay is formulated with a queuing model. Evaluation results show that the proposed algorithm guarantees desired performance improvements.

Electrical and mechanical failures in electric vehicles (EVs) during passenger operation cause significant operational losses and elevated energy consumption, amplifying range anxiety. To address this issue, we utilized 250,000 rows of real-time data from electric trolleybuses operating in Turkiye to develop a robust artificial intelligence (AI)-based optimization model for failure mitigation. Initially, Tri layered Neural Network (TNN) was employed to create a predictive function for electrical and mechanical failures, followed by comparative analyses across six optimization algorithms widely adopted in failure prediction studies. Among these, the Developed Honey Badger Algorithm with AI Approach (DHBA) emerged as the most effective, achieving a predictive accuracy improvement of 15 % over the standard Honey Badger Algorithm (HBA). The DHBA incorporates a Dynamic Fitness-Distance Balance (DFDB) mechanism and a novel spiral motion feature to enhance search precision, leading to the DHBA-FPM (Developed-Honey Badger Algorithm - Failure Prediction Model). The final DHBA-FPM model was applied to the 10 highest-density bus routes in Turkiye to predict and optimize failures. Results indicate that applying the DHBA-FPM model across these routes yielded a 3.96 % average range increase in EVs, extending the total range by approximately 79,200 km annually. It can be concluded that the model could prevent the release of 238.7 tons/year of CO2, NO, and NO2 emissions through its potential to improve both the operational efficiency and sustainability of EVs in public transit networks.

By considering the limited energy of Internet of Things (IoT) devices. We take the resource allocation to guarantee the stringent Quality of Service (QoS) depending on the joint optimization of power control and finite blocklength of channel. To achieve large volumes of arrival rates, we propose Adversarial Training based Generative Adversarial Networks (AT-GANs), which utilize a significant number of extreme events to provide high reliability and adjust real data in real-time. Simulation results show that Deep-Reinforcement Learning (Deep-RL) for AT-GAN could eliminate the transient training time. As a result, the AT-GAN keeps the reliability higher than 99.9999%.