• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2021.05a
• /
• pp.25-27
• /
• 2021

With the recent development of deep learning technology, computer vision-based AI technologies have been studied to analyze the abnormal behavior of objects in image information acquired through CCTV cameras. There are many cases where surveillance cameras are installed in dangerous areas or security areas for crime prevention and surveillance. For this reason, companies are conducting studies to determine major situations such as intrusion, roaming, falls, and assault in the surveillance camera environment. In this paper, we propose a real-time abnormal behavior analysis algorithm using object detection and tracking method.

• Journal of Korea Multimedia Society
• /
• v.12 no.1
• /
• pp.97-106
• /
• 2009

WLAN(wireless local area network)-based positioning is the most attractive because it does not require any special equipments dedicated for positioning even though it is less accurate than the other strategies. Applying our WLAN-based decision tree method for indoor positioning, we obtained pedestrian's tracks, and performed stochastic confidence tests on the tracks in order to validate them.

• Journal of Positioning, Navigation, and Timing
• /
• v.9 no.3
• /
• pp.169-173
• /
• 2020

Location-based services are used as core services in various fields. In particular, in the field of public services such as emergency rescue, accurate location estimation technology is very important. Recently, the technology of tracking the location of self-isolation subjects for COVID-19 has become a major issue. Therefore, location estimation technology using personal smart devices is being studied in various ways, and the most widely used method is to use GPS. Other representative methods are using Wi-Fi, Pedestrian Dead Reckoning (PDR), Bluetooth Low Energy (BLE) beacons, and LTE signals. In this paper, we introduced a positioning technology using deep learning based on LTE Channel State Information-Reference Signal (CSI-RS) data, and confirmed the possibility through an outdoor location estimation experiment using a commercial LTE signal.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2015.10a
• /
• pp.1302-1304
• /
• 2015

컴퓨터 비전에서 다수의 보행자 궤적을 생성하는 문제는 여전히 어려운 문제이다. 전경에서 추출된 보행자 윤곽은 음영과 밝기 등의 문제로 윤곽이 명확하지 않고, 보행자들이 서로 다른 방향으로 움직이며 상호작용을 한다. 이로 인해 보행자를 식별하고 궤적을 생성하기에는 다소 어려움이 있다. 우리는 의사결정 트리를 사용하여 보행자 영역의 병합과 분할 상황을 개별 분리된 보행자로 검출한다. 검출된 개별 보행자는 점 대응 알고리즘으로 각 보행자의 궤적을 생성한다. 우리는 수정된 $A^*$ 검색 알고리즘으로 새로운 휴리스틱 점 대응 알고리즘을 소개한다. 우리의 실험은 PETS2010 데이터 세트로 구현되고 실험했다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.06a
• /
• pp.246-248
• /
• 2019

최근 딥 러닝을 이용한 방법들이 이미지 분류에서 뛰어난 성능을 보임에 따라, 컴퓨터 비전의 중요한 문제 중 하나인 이미지 검색에도 이를 활용하고 있다. 특히, 이미지 검색에 사용할 수 있는 이미지 기술자 (Image descriptor)를 깊은 신경망 구조의 일부분인 Fully-connected layer에서 추출하여 사용하는 방법들이 제시되고 있고, 이를 위해 알맞은 목적함수를 설계하여 깊은 신경망을 학습하는 것이 중요해지고 있다. 딥 러닝을 통해 얻은 이미지 기술자는 실수형 데이터로서, 한 장의 이미지를 수치화하여 표현하는 데 많은 메모리를 소모하게 된다. 이를 보완하기 위해 이미지 기술자를 작은 용량의 이진코드로 mapping 하는 해싱 (hashing) 이라는 과정이 필수적이나 이에 따른 한계점이 발생한다. 본 연구에서는 실수형 데이터가 갖는 거리 계산에서의 이점과 이진코드의 장점을 동시에 살릴 수 있는 Product Quantization 방식의 이미지 검색 방법을 이용하여 한계점을 극복하였다. 우리는 제안한 방법을 얼굴 이미지 데이터 셋에 실험하였고 기존 방식보다 뛰어난 성능을 보이는 것을 확인할 수 있었다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2018.06a
• /
• pp.79-81
• /
• 2018

최근 딥 러닝의 발전과 함께 보행자 검출 기술의 성능이 발전하면서 다양한 분야에서 응용되고 있다. 영상 내 보행자의 위치나 움직임을 파악함으로써 위험 지역이나 보안 지역에 접근하는 보행자를 찾아낼 수 있다. 일반적인 딥 러닝 기반의 물체 검출기는 멀리 있는 보행자와 같은 작은 물체를 검출 하는 데에 적합하지 않다. 또, 검출을 수행하기 위해서 큰 계산량을 필요로 하기 때문에, 동영상의 매 프레임 마다 수행하기 부적합 하다는 단점이 있다. 본 논문에서는 작은 물체도 잘 검출할 수 있도록 기존 YOLO 네트워크의 구조를 변경하고, 보행자 데이터를 이용하여 추가로 학습함으로써 보행자를 검출하는 성능을 증가시켰다. 그리고 검출한 보행자들에 대해 추적 기법을 이용함으로써, 동영상의 매 프레임 마다 검출을 수행하는 것을 피할 수 있도록 하였다. 실제로 DukeMTMC Dataset을 이용하여 실험을 해본 결과, YOLO 네트워크의 구조를 변경하고 추가 학습을 함으로써 검출 정확도가 개선되는 것을 확인할 수 있었다. 또, 추적 기법을 이용했을 때, 성능이 크게 떨어지지 않으면서 검출 속도를 개선할 수 있는 것을 확인할 수 있었다.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.6 no.1
• /
• pp.1-5
• /
• 2013

The number of pedestrians is considered essential information which can be used to control a person who makes a entrance or a exit into a building. The number of pedestrians, also, can be used to help to manage pedestrian traffic and the volume of pedestrian flow within the building. Due to the fact there is incorrect detection by occluded, shadows, and illumination, however, difficulty can arise in existing system which is for detection and counts of a person who makes a entrance or a exit into a building. In this paper, it is minimized that the change of illumination and the effect of shadow through the transmitted image from camera which is created and processed with great adaptability. The accuracy of the calculations can be increase as well by using Kalman Filter and Mean-Shift Algorithm in order to avoid overlapped counts. As a result of the test, it is proved that the count method that shows the accuracy of 95.4% should be effective for detection and counts.

• Journal of Digital Convergence
• /
• v.15 no.2
• /
• pp.173-181
• /
• 2017

In this paper, we have studied tracking as a training stage of considering the position and the scale of a person given its previous position, scale, as well as next and forward image fraction. Unlike other learning methods, CNN is thereby learning combines both time and spatial features from the image for the two consecutive frames. We introduce multiple path ways in CNN to better fuse local and global information. A creative shift-variant CNN architecture is designed so as to alleviate the drift problem when the distracting objects are similar to the target in cluttered environment. Furthermore, we employ CNNs to estimate the scale through the accurate localization of some key points. These techniques are object-independent so that the proposed method can be applied to track other types of object. The capability of the tracker of handling complex situations is demonstrated in many testing sequences. The accuracy of the SVM classifier using the features learnt by the CNN is equivalent to the accuracy of the CNN. This fact confirms the importance of automatically optimized features. However, the computation time for the classification of a person using the convolutional neural network classifier is less than approximately 1/40 of the SVM computation time, regardless of the type of the used features.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.8 no.2
• /
• pp.149-157
• /
• 2009

Tracking technologies which provide real-time and customized information through various information collecting and processing for pedestrians who use traffic connective and transferring center have been being examined. However some problems are caused due to the wide-range positioning error for some services as device installation and service place. It is also difficult to be applied to traffic linkage and transfer services because many situations can be barren. In the testbed, Gwangmyoung Station, we got some results in bad conditions such as a lot of steel construction and another communication device. Practically, conditions of the place which will be built can be worse than Gwangmyoung station. Therefore, we researched suitable Location correction algorithm as a method for accuracy to traffic connective and transferring system. And its algorithm is designed through grid coordinates, map-matching, modeling coordinates and Kalman filtering and is being implemented continuously. Also preparing for optimization of various transferring center model, we designed for simulator type algorithm what is available for deciding algorithm factor.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.22 no.5
• /
• pp.43-52
• /
• 2023

Traffic accidents among children have been decreasing after the installation of yellow carpets. However, the explanatory power of the causal relationship between yellow carpet installation and traffic accidents is still insufficient. The yellow carpet effect was analyzed in greater depth using virtual reality (VR) simulation experiments in various situation that could not be evaluated in existing actual vehicle research studies due to difficulties or risks in implementation. A target site where an actual yellow carpet was installed was selected and, implemented into a virtual environment. Subjects were made to, were gaze measurement equipment and ride the simulator. The visual/driving behavior before and after yellow carpet installation was compared, and a t-test analysis was performed for statistical verification. All the results were found to be statistically significant.