• Journal of IKEEE
• /
• v.23 no.4
• /
• pp.1302-1308
• /
• 2019

This paper presents an approach for classifying image made by acquired position data from a 2D Lidar sensor with a convolutional neural network (CNN). Lidar sensor has been widely used for unmanned devices owing to advantages in term of data accuracy, robustness against geometry distortion and light variations. A CNN algorithm consists of one or more convolutional and pooling layers and has shown a satisfactory performance for image classification. In this paper, different types of CNN architectures based on training methods, Gradient Descent(GD) and Levenberg-arquardt(LM), are implemented. The LM method has two types based on the frequency of approximating Hessian matrix, one of the factors to update training parameters. Simulation results of the LM algorithms show better classification performance of the image data than that of the GD algorithm. In addition, the LM algorithm with more frequent Hessian matrix approximation shows a smaller error than the other type of LM algorithm.

• The Bulletin of The Korean Astronomical Society
• /
• v.41 no.2
• /
• pp.48.2-48.2
• /
• 2016

The wide-field and the round-the clock operation capabilities of the KMTNet enables the discovery, astrometry and follow-up physical characterization of asteroids and comets in a most efficient way. We collectively refer to the team members, partner organizations, the dedicated software subsystem, the computing facility and research activities as Deep Ecliptic Patrol of the Southern Sky (DEEP-South). Most of the telescope time for DEEP-South is devoted to targeted photometry of Near Earth Asteroids (NEAs) to push up the number of the population with known physical properties from several percent to several dozens of percent, in the long run. We primarily adopt Johnson R-band for lightcurve study, while we employ BVI filters for taxonomic classification and detection of any possible color variations of an object at the same time. In this presentation, the progress and new findings since the last KAS meeting will be outlined. We report DEEP-South preliminary lightcurves of several dozens of NEAs obtained at three KMTNet stations during the first year runs. We also present a physical model of asteroid (5247) Krylov, the very first Non principal Axis (NPA) rotator that has been confirmed in the main belt (MB). A new asteroid taxonomic classification scheme will be introduced with an emphasis on its utility in the LSST era. The progress on the current version of automated mover detection software will also be summarized.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.23 no.11
• /
• pp.1357-1363
• /
• 2019

In this paper, we compares the performance of the gredient descent optimizers of the Faster Region-based Convolutional Neural Network (R-CNN) model for the chromosome object detection in digital images composed of human metaphase chromosomes. In faster R-CNN, the gradient descent optimizer is used to minimize the objective function of the region proposal network (RPN) module and the classification score and bounding box regression blocks. The gradient descent optimizer. Through performance comparisons among these four gradient descent optimizers in our experiments, we found that the Adamax optimizer could achieve the mean average precision (mAP) of about 52% when considering faster R-CNN with a base network, VGG16. In case of faster R-CNN with a base network, ResNet50, the Adadelta optimizer could achieve the mAP of about 58%.

• International conference on construction engineering and project management
• /
• 2024.07a
• /
• pp.423-430
• /
• 2024

The task of vision safety monitoring in construction environments presents a formidable challenge, owing to the dynamic and heterogeneous nature of these settings. Despite the advancements in artificial intelligence, the nuanced analysis of small or tiny personal protective equipment (PPE) remains a complex endeavor. In response to this challenge, this paper introduces an innovative safety monitoring system, specifically designed to enhance the safety monitoring of working both at ground level and at elevated heights. This novel system integrates a suite of sophisticated technologies: instance segmentation, shape classification, object tracking, a visualization report, and a real-time notification module. Collectively, these components coalesce to deliver a safety monitoring solution, ensuring a higher standard of protection for construction workers. The experimental results…..

• Journal of Ubiquitous Convergence Technology
• /
• v.1 no.1
• /
• pp.3-8
• /
• 2007

This paper presents a set of techniques used in a real-time visual surveillance system. The system is implemented on a low-cost embedded DSP platform that is designed to work with stationary video sources. It consists of detection, a tracking and a classification module. The detector uses a statistical method to establish the background model and extract the foreground pixels. These pixels are grouped into blobs which are classified into single person, people in a group and other objects by the dynamic periodicity analysis. The tracking module uses mean shift algorithm to locate the target position. The system aims to control the human density in the surveilled scene and detect what happens abnormally. The major advantage of this system is the real-time capability and it only requires a video stream without other additional sensors. We evaluate the system in the real application, for example monitoring the subway entrance and the building hall, and the results prove the system's superior performance.

• Journal of KIISE:Computing Practices and Letters
• /
• v.15 no.6
• /
• pp.459-463
• /
• 2009

Background modeling and motion detection is the one of the most significant real time video processing technique. Until now, many researches are conducted into the topic but it still needs much time for robustness. It is more important when other algorithms are used together such as object tracking, classification or behavior understanding. In this paper, we propose efficient multi-modal background modeling methods which can be understood as simplified learning method of Gaussian mixture model. We present its validity using numerical methods and experimentally show detecting performance.

• International journal of advanced smart convergence
• /
• v.6 no.4
• /
• pp.73-79
• /
• 2017

The Convolutional Neural Network (CNN) has shown an excellent performance in computer vision task. Applications of CNN include image classification, object detection in images, autonomous driving, etc. This paper will evaluate the performance of CNN model with ReLU and SELU as activation function. The evaluation will be performed on four different choices of hyperparameter which are initialization method, network configuration, optimization technique, and regularization. We did experiment on each choice of hyperparameter and show how it influences the network convergence and test accuracy. In this experiment, we also discover performance improvement when using SELU as activation function over ReLU.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2018.10a
• /
• pp.441-443
• /
• 2018

This paper deals with the analysis of leg motion using RGB image. We used RGB image as gait analysis element by using BMC(Background Model Challenge) method and by using combining object recognition segmentation algorithm and attitude detection algorithm. It is considered that gait analysis incorporating image can be used as a parameter for classification of gait pattern recognition and abnormal gait.

• Journal of Korea Multimedia Society
• /
• v.23 no.7
• /
• pp.883-890
• /
• 2020

In this paper, we propose a variational expectation-maximization algorithm that computes posterior probabilities from Latent Dirichlet Allocation (LDA) model. The algorithm approximates the intractable posterior distribution of a document term matrix generated from a corpus made up by 50 papers. It approximates the posterior by searching the local optima using lower bound of the true posterior distribution. Moreover, it maximizes the lower bound of the log-likelihood of the true posterior by minimizing the relative entropy of the prior and the posterior distribution known as KL-Divergence. The experimental results indicate that documents clustered to image classification and segmentation are correlated at 0.79 while those clustered to object detection and image segmentation are highly correlated at 0.96. The proposed variational inference algorithm performs efficiently and faster than Gibbs sampling at a computational time of 0.029s.

• Proceedings of the KSRS Conference
• /
• 2008.10a
• /
• pp.72-75
• /
• 2008

Image segmentation techniques becoming increasingly important in the field of remote sensing image analysis in areas such as object oriented image classification. This paper presents a new method for image segmentation in High Resolution Remote Sensing Image based on Seeded Region Growing (SRG) and Edge Information. Firstly, multi-spectral edge detection was done using an entropy operator in pan-sharpened QuickBird imagery. Then, the initial seeds were automatically selected from the obtained edge map. After automatic selection of significant seeds, an initial segmentation was achieved by applying SRG. Finally the region merging process, using region adjacency graph (RAG), was carried out to get the final segmentation result. Experimental results demonstrated that the proposed method has good potential for application in the segmentation of high resolution satellite images.