• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.14 no.4
• /
• pp.239-246
• /
• 2021

In the field of computer graphics and HCI, there are many studies on systems that create characters and interact naturally. Such studies have focused on the user's response to the user's behavior, and the study of the character's behavior to elicit positive emotions from the user remains a difficult problem. In this paper, we develop a prototype of an interaction system to elicit positive emotions from users according to the movement of virtual characters using artificial intelligence technology. The proposed system is divided into face recognition and motion generation of a virtual character. A depth camera is used for face recognition, and the recognized data is transferred to motion generation. We use imitation learning as a learning model. In motion generation, random actions are performed according to the first user's facial expression data, and actions that the user can elicit positive emotions are learned through continuous imitation learning.

• Smart Media Journal
• /
• v.11 no.3
• /
• pp.90-101
• /
• 2022

Buses are a popular means of transportation. As such, thorough preparation is needed for passenger safety management. However, the safety system is insufficient because there are accidents such as a death accident occurred when the bus departed without recognizing the elderly approaching to get on in 2018. There is a safety system that prevents pinching accidents through sensors on the back door stairs, but such a system does not prevent accidents that occur in the process of getting on and off like the above accident. If it is possible to predict the intention of bus passengers to get on and off, it will help to develop a safety system to prevent such accidents. However, studies predicting the intention of passengers to get on and off are insufficient. Therefore, in this paper, we propose a 1×1 CNN3D-based getting on and off intention prediction model using skeleton keypoints of passengers extracted from the camera image attached to the bus through UDP-Pose. The proposed model shows approximately 1~2% higher accuracy than the RNN and LSTM models in predicting passenger's getting on and off intentions.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.18 no.5
• /
• pp.877-884
• /
• 2023

In this study, when another vehicle is parked in a designated space where a personal vehicle can park and a defined personal use time, an ultrasonic object recognition sensor is used to determine vehicle entry, and a camera sensor recognizes a license plate. If the vehicle is not recognized by the individual vehicle owner, the "private parking lot operation block" of the application server receives the individual phone number based on the National Police Agency's Vehicle Number Information Inquiry Open API. Afterwards, when parking is processed, the non-right holder receives the approval of the parking right holder, parks for the recognized time, and deposits the parking fee into the public account of the city hall. Through this study, it was possible to find an operation processing method that can most effectively manage parking in the city center in a private parking space recognized by the city hall.

• KIPS Transactions on Software and Data Engineering
• /
• v.12 no.12
• /
• pp.525-534
• /
• 2023

Gas leak detection system is a key to minimize the loss of life due to the explosiveness and toxicity of gas. Most of the leak detection systems detect by gas sensors or thermal imaging cameras. To improve the performance of gas leak detection system using single-modal methods, the paper propose multimodal approach to gas sensor data and thermal camera data in developing a gas type identification model. MultimodalGasData, a multimodal open-dataset, is used to compare the performance of the four models developed through multimodal approach to gas sensors and thermal cameras with existing models. As a result, 1D CNN and GasNet models show the highest performance of 96.3% and 96.4%. The performance of the combined early fusion model of 1D CNN and GasNet reached 99.3%, 3.3% higher than the existing model. We hoped that further damage caused by gas leaks can be minimized through the gas leak detection system proposed in the study.

• Journal of Broadcast Engineering
• /
• v.25 no.2
• /
• pp.200-207
• /
• 2020

Recently, deep neural networks such as CNN are showing excellent performance in various fields such as image classification, object recognition, visual quality enhancement, etc. However, as the model size and computational complexity of deep learning models for most applications increases, it is hard to apply neural networks to IoT and mobile environments. Therefore, neural network compression algorithms for reducing the model size while keeping the performance have been being studied. In this paper, we apply few compression methods to CNN models and evaluate their performances in the embedded environment. For evaluate the performance, the classification performance and inference time of the original CNN models and the compressed CNN models on the image inputted by the camera are evaluated in the embedded board equipped with QCS605, which is a customized AI chip. In this paper, a few CNN models of MobileNetV2, ResNet50, and VGG-16 are compressed by applying the methods of pruning and matrix decomposition. The experimental results show that the compressed models give not only the model size reduction of 1.3~11.2 times at a classification performance loss of less than 2% compared to the original model, but also the inference time reduction of 1.2~2.21 times, and the memory reduction of 1.2~3.8 times in the embedded board.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.22 no.6
• /
• pp.114-123
• /
• 2023

With the development of computer vision technology, video sensors such as CCTV are detecting incident. However, most of the current incident have been detected based on existing fixed imaging equipment. Accordingly, there has been a limit to the detection of incident in shaded areas where the image range of fixed equipment is not reached. With the recent development of edge-computing technology, real-time analysis of mobile image information has become possible. The purpose of this study is to evaluate the possibility of detecting expressway emergencies by introducing computer vision technology to dash cam. To this end, annotation data was constructed based on 4,388 dash cam still frame data collected by the Korea Expressway Corporation and analyzed using the YOLO algorithm. As a result of the analysis, the prediction accuracy of all objects was over 70%, and the precision of traffic accidents was about 85%. In addition, in the case of mAP(mean Average Precision), it was 0.769, and when looking at AP(Average Precision) for each object, traffic accidents were the highest at 0.904, and debris were the lowest at 0.629.

• The Journal of Korean Medicine
• /
• v.27 no.3 s.67
• /
• pp.187-200
• /
• 2006

Objectives: Acupuncture has been applied in Asia for thousands of years, especially to rehabilitation after stroke. It has been reported that acupuncture increased cerebral blood supply and stimulated the functional activity of brain nerve cells shown by using brain imaging techniques. This study was to evaluate the effect of GB 34-GB 39 electro-acupuncture (EA) on regional cerebral blood flow (rCBF) in stroke patients and normal volunteers using single photon emission computed tomography (SPECT). Methods: The study procedure was divided into two parts: patients and volunteers studies. For the patients study, ten ischemic stroke patients (3 males, 7 females, mean age $68.5{\pm}8.9$ years old) were selected. Baseline brain SPECT was done with triple head gamma camera (MultiSPECT3, Siemens, USA) after intravenous administration of 1,110 MBq of $^{99m}Tc-ECD$. Fifteen-minute EA at GB 34 and GB 39 were applied on the affected limb. The same dose of $^{99m}Tc-ECD$ was injected during the EA, and the second set of SPECT images wasobtained. Using the computer software (ICON 7.1, Siemens, USA), 3 SPECT slices (upper, middle, lower) surrounding the brain lesion were selected and each slice was divided into 10-16 brain regions. Asymmetry indexes (AI) were analyzed in each brain region. We regarded over 10% changes of AI between before and after EA as significance. For the volunteers study, 10 healthy human volunteers (5 males, 5 females, mean age $28.1{\pm}6$ years old) were selected. In the resting state, $^{99m}Tc-ECD$ brain SPECT scans were performed. On the 7th day after the resting examination, 15 minute EA was applied at GB 34 and GB 39 on the right side of the subjects. Immediately after EA, the second SPECT images were obtained inthe same manner as the resting state. Significant increases and decreases of rCBF after EA were estimated by comparing their SPECT images with those of the resting state using paired t statistics at every voxel, which were analyzed by statistical parametric mapping with a threshold of p = 0.01, uncorrected (extent threshold: k=100 voxels). Results: In stroke patients, six of the eight (75%) had significantly increased perfusion in post-acupuncture scans compared to their baseline state. In normal volunteers, GB 34-GB GB EA increased rCBF in both hemispheres including right ventral posterior cingulate (Brodmann area (BA) 23), left superior temporal, anterior transverse temporal (BA 22, 41), left parastriate, peristriate (BA 18, 19), right occipitotemporal, angular (BA 37, 39), left rostral postcentral, caudal postcentral and preparietal (BA 2, 3, 5). However GB 34-GB 39 EA decreased rCBF in the right hemisphere including triangular and middle frontal lobes. Conclusions: The results demonstrated that OB 34-GB 39 EA increased cerebral perfusion in ischemic stroke patients and increased rCBF grossly in temporal lobes of normal volunteers. It is also suggested that there may be a correlation between the GB meridian and the territory of the middle cerebral artery.