• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.9 no.1
• /
• pp.9-18
• /
• 2010

Our country has so large mountain area that the tunnel construction is inevitable and the need of incident detection that provides safe management of tunnels is increasing. In this paper, we suggest a tunnel incident detection system using computer vision techniques, which can detect the incidents in a tunnel and provides the information to the tunnel administrative office in order to help safe tunnel operation. The suggested system enhances the processing speed by using simple processing algorithm such as image subtraction, and ensures the accuracy of the system by focused on the incident detection itself rather than its classification. The system is also cost effective because the video data from 4 cameras can be simultaneously analyzed in a single PC-based system. Our system can be easily extended because the PC-based analyzer can be increased according to the number of cameras in a tunnel. Also our web-based structure is useful to connect the other remotely located tunnel incident systems to obtain interoperability between tunnels. Through the experiments the system has successfully detected the incidents in real time including dropped luggage, stoped car, traffic congestion, man walker or bicycle, smoke or fire, reverse driving, etc.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.20 no.5
• /
• pp.9-13
• /
• 2020

Sasang Constitutional Medicine is a traditional Korean medicine optimized for personalized medicine, and despite its effective clinical efficacy, the inaccuracy of constitutional diagnosis has been pointed out as a limitation. To improve the accuracy, a constitutional analysis algorithm based on quantitative data was developed. In this study, a constitutional analysis experience system applied with the algorithm was developed and repeatability was evaluated. The system analyzes the constitution of the experiencer by collecting front and side facial images, audio, and questionnaire and calculating the integrated constitution probability value. To evaluate the repeatability of the probability values of the system was performed five times each for three people, and the coefficient of variation was 4.778%, indicating that the repeatability was sufficient. The system could contribute to the promotion of the awareness of Sasang medicine.

• Journal of IKEEE
• /
• v.24 no.1
• /
• pp.225-231
• /
• 2020

In this paper, we propose a method to accelerate BWN based on FPGA with limited resources for embedded system. Because of the limited number of logic elements available, a single computing unit capable of handling Conv-layer, FC-layer of various sizes must be designed and reused. Also, if the input feature map can not be parallel processed at one time, the output must be calculated by reading the inputs several times. Since the number of available BRAM modules is limited, the number of data bits in the BWN accelerator must be minimized. The image classification processing time of the BWN accelerator is superior when compared with a embedded CPU and is faster than a desktop PC and 50% slower than a GPU system. Since the BWN accelerator uses a slow clock of 50MHz, it can be seen that the BWN accelerator is advantageous in performance versus power.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.13 no.10
• /
• pp.2180-2188
• /
• 2009

This paper is study on real-time hand gesture recognition system based on vision for intelligent robot control. We are proposed a recognition system using PCA and BP algorithm. Recognition of hand gestures consists of two steps which are preprocessing step using PCA algorithm and classification step using BP algorithm. The PCA algorithm is a technique used to reduce multidimensional data sets to lower dimensions for effective analysis. In our simulation, the PCA is applied to calculate feature projection vectors for the image of a given hand. The BP algorithm is capable of doing parallel distributed processing and expedite processing since it take parallel structure. The BP algorithm recognized in real time hand gestures by self learning of trained eigen hand gesture. The proposed PCA and BP algorithm show improvement on the recognition compared to PCA algorithm.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2014.11a
• /
• pp.798-801
• /
• 2014

여행 정보 시스템(ATIS), 교통 관리 시스템 (ITS) 등 궤적 기반 서비스에서, 서비스 품질을 향상시키기 위해서는 주어진 궤적 질의에 대한 정확한 주행시간을 예측하는 것이 필수적이다. 이를 위한 대표적인 공간 데이터 분석 기법으로는 데이터 분류에서 높은 정확도를 보장하는 규칙 기반 분류화 기법이 존재한다. 그러나 기존 규칙 기반 분류화 기법은 단일 컴퓨터 환경만을 고려하기 때문에, 대용량 공간 데이터 처리에 적합하지 않은 문제점이 존재한다. 이를 해결하기 위해, 본 연구에서는 맵리듀스 환경에서 규칙 기반 분류화를 이용한 궤적 데이터 주행 시간 예측 알고리즘을 개발하고자 한다. 제안하는 알고리즘은 첫째, 맵리듀스를 이용하여 대용량 공간 데이터를 병렬적으로 분석함으로써, 활용도 높은 궤적 데이터 규칙을 생성한다. 이를 통해 대용량 공간 데이터 기반의 규칙 생성 시간을 감소시킨다. 둘째, 그리드 구조 기반의 지도 데이터 분할을 통해, 사용자 질의처리 시 탐색 성능을 향상시킨다. 즉, 주행 시간 예측을 위한 규칙 그룹을 탐색 시 질의를 포함하는 그리드 셀만을 탐색하기 때문에, 질의처리 성능이 향상된다. 마지막으로 맵리듀스 구조에 적합한 질의처리 알고리즘을 설계하여, 효율적인 병렬 질의처리를 지원한다. 이를 위해 맵 함수에서는 선정된 그리드 셀에 대해, 질의에 포함된 도로 구간에서의 주행 시간을 병렬적으로 측정한다. 아울러 리듀스 함수에서는 출발 시간 및 구간별 주행 시간을 바탕으로 맵 함수의 결과를 병합함으로써, 최종 결과를 생성한다. 이를 통해 공간 빅데이터 분석을 통한 주행 시간 예측 기법의 처리 시간 및 결과 정확도를 향상시킨다.

• The Transactions of the Korea Information Processing Society
• /
• v.5 no.8
• /
• pp.2061-2074
• /
• 1998

In pattern recognition and image analysis upplications, a graph is a useful tool for complex obect representation and recognition. However it takes much time to pair proper nodes between the prototype graph and an input data graph. Futhermore it is difficult to decide whether the two graphs in a class are the same hecause real images are degradd in general by noise and other distortions. In this paper we propose a matching algorithm using a matrix. The matrix is suiable for simple and easily understood representation and enables the ordering and matching process to be convenient due to its predefined matrix manipulation. The nodes which constitute a gaph are ordered in the matrix by their geometrical positions and this makes it possible to save much comparison time for finding proper node pairs. for the classification, we defined a distance measure thatreflects the symbo's structural aspect that is the sum of the mode distance and the relation distance; the fornet is from the parameters describing the node shapes, the latter from the relations with othes node in the matrix. We also introduced a subdivision operation to compensate node merging which is mainly due t the prepreocessing error. The proposed method is applied to the recognition of musteal symbols and the result is given. The result shows that almost all, except heavily degraded symbols are recognized, and the recognition rate is approximately 95 percent.

• Journal of the Korean Society of Environmental Restoration Technology
• /
• v.24 no.2
• /
• pp.1-17
• /
• 2021

As population concentration, high-rise, and high-density progresses amid rapid industrialization and urbanization in Korea, green areas in the city have decreased significantly and the natural environment has deteriorated day by day due to various development projects. In order to cope with these social changes, many studies related to park and green areas have been conducted, but in the case of Gwangju-si, only studies on green areas such as vegetation structures, street facilities, and recreation forests have been conducted on park green areas. The situation is not losing. Therefore, in this study, a survey was conducted on Gwangju citizens to understand the usage behavior and satisfaction of park green areas, and to provide basic data for future park development and reorganization. As a result, first, to induce a variety of static and dynamic activities in order to increase the park satisfaction of Gwangju citizens, it is possible to create a park that has not been formed as soon as possible, and by planning programs with various themes. Second, it is necessary to expand green space services in marginalized areas by creating water streams, flower pots, and rooftop greening around the lives of Gwangju citizens, and create park green spaces with themes using cultural and historical resources for each administrative district. Third, in order to solve the dry landscape in some urban areas, it is necessary to establish a three-dimensional landscape plan by creating a variety of green areas such as potted plants, green areas, and street trees, and by giving functions of park green areas such as expanding leisure and convenience facilities.

• Resources Recycling
• /
• v.28 no.2
• /
• pp.23-30
• /
• 2019

An color sorting technique was introduced to recycle End-of-life automobile shredded bumpers. The color sorting is a innovate method of separating the differences in the color of materials which are difficult to separate in gravity and size classification by using a camera and an image process technique. Experiments were planned and optimal conditions were derived by applying BBD (Box-Behnken Design) in the reaction surface method. The effects of color sensitivity, feed rate and sample size were analyzed, and a second-order reaction model was obtained based on the analysis of regression and statistical methods and $R^2$ and p-value were 99.56% and < 0.001. Optimum recovery was 94.1% under the conditions of color sensitivity, feed rate and particle size of 32%, 200 kg/h, and 33 mm respectively. The recovery of actual experiment was 93.8%. The experimental data agreed well with the predicted value and confirmed that the model was appropriate.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.14 no.5
• /
• pp.277-286
• /
• 2019

In this paper, we propose an efficient hyperplane generation technique to classify human activity from combination of events and sequence information obtained from multiple-event sensors. By generating hyperplane efficiently, our machine learning algorithm classify with less memory and run time than the LSVM (Linear Support Vector Machine) for embedded system. Because the fact that light weight and high speed algorithm is one of the most critical issue in the IoT, the study can be applied to smart home to predict human activity and provide related services. Our approach is based on reducing numbers of hyperplanes and utilizing robust string comparing algorithm. The proposed method results in reduction of memory consumption compared to the conventional ML (Machine Learning) algorithms; 252 times to LSVM and 34,033 times to LSTM (Long Short-Term Memory), although accuracy is decreased slightly. Thus our method showed outstanding performance on accuracy per hyperplane; 240 times to LSVM and 30,520 times to LSTM. The binarized image is then divided into groups, where each groups are converted to binary number, in order to reduce the number of comparison done in runtime process. The binary numbers are then converted to string. The test data is evaluated by converting to string and measuring similarity between hyperplanes using Levenshtein algorithm, which is a robust dynamic string comparing algorithm. This technique reduces runtime and enables the proposed algorithm to become 27% faster than LSVM, and 90% faster than LSTM.

• Imaging Science in Dentistry
• /
• v.52 no.2
• /
• pp.219-224
• /
• 2022

Purpose: This study aimed to evaluate the performance of transfer learning in a deep convolutional neural network for classifying implant fixtures. Materials and Methods: Periapical radiographs of implant fixtures obtained using the Superline (Dentium Co. Ltd., Seoul, Korea), TS III(Osstem Implant Co. Ltd., Seoul, Korea), and Bone Level Implant(Institut Straumann AG, Basel, Switzerland) systems were selected from patients who underwent dental implant treatment. All 355 implant fixtures comprised the total dataset and were annotated with the name of the system. The total dataset was split into a training dataset and a test dataset at a ratio of 8 to 2, respectively. YOLOv3 (You Only Look Once version 3, available at https://pjreddie.com/darknet/yolo/), a deep convolutional neural network that has been pretrained with a large image dataset of objects, was used to train the model to classify fixtures in periapical images, in a process called transfer learning. This network was trained with the training dataset for 100, 200, and 300 epochs. Using the test dataset, the performance of the network was evaluated in terms of sensitivity, specificity, and accuracy. Results: When YOLOv3 was trained for 200 epochs, the sensitivity, specificity, accuracy, and confidence score were the highest for all systems, with overall results of 94.4%, 97.9%, 96.7%, and 0.75, respectively. The network showed the best performance in classifying Bone Level Implant fixtures, with 100.0% sensitivity, specificity, and accuracy. Conclusion: Through transfer learning, high performance could be achieved with YOLOv3, even using a small amount of data.