• Journal of KIISE:Software and Applications
• /
• v.32 no.11
• /
• pp.1121-1135
• /
• 2005

This paper proposes a moving object extraction method using the contrast map and salient points. In order to make the contrast map, we generate three-feature maps such as luminance map, color map and directional map and extract salient points from an image. By using these features, we can decide the Attention Window(AW) location easily The purpose of the AW is to remove the useless regions in the image such as background as well as to reduce the amount of image processing. To create the exact location and flexible size of the AW, we use motion feature instead of pre-assumptions or heuristic parameters. After determining of the AW, we find the difference of edge to inner area from the AW. Then, we can extract horizontal candidate region and vortical candidate region. After finding both horizontal and vertical candidates, intersection regions through logical AND operation are further processed by morphological operations. The proposed algorithm has been applied to many video sequences which have static background like surveillance type of video sequences. The moving object was quite well segmented with accurate boundaries.

• Smart Structures and Systems
• /
• v.29 no.1
• /
• pp.1-16
• /
• 2022

Despite recent breakthroughs in deep learning and computer vision fields, the pixel-wise identification of tiny objects in high-resolution images with complex disturbances remains challenging. This study proposes a modified U-net for tiny crack segmentation in real-world steel-box-girder bridges. The modified U-net adopts the common U-net framework and a novel Self-Attention-Self-Adaption (SASA) neuron as the fundamental computing element. The Self-Attention module applies softmax and gate operations to obtain the attention vector. It enables the neuron to focus on the most significant receptive fields when processing large-scale feature maps. The Self-Adaption module consists of a multiplayer perceptron subnet and achieves deeper feature extraction inside a single neuron. For data augmentation, a grid-based crack random elastic deformation (CRED) algorithm is designed to enrich the diversities and irregular shapes of distributed cracks. Grid-based uniform control nodes are first set on both input images and binary labels, random offsets are then employed on these control nodes, and bilinear interpolation is performed for the rest pixels. The proposed SASA neuron and CRED algorithm are simultaneously deployed to train the modified U-net. 200 raw images with a high resolution of 4928 × 3264 are collected, 160 for training and the rest 40 for the test. 512 × 512 patches are generated from the original images by a sliding window with an overlap of 256 as inputs. Results show that the average IoU between the recognized and ground-truth cracks reaches 0.409, which is 29.8% higher than the regular U-net. A five-fold cross-validation study is performed to verify that the proposed method is robust to different training and test images. Ablation experiments further demonstrate the effectiveness of the proposed SASA neuron and CRED algorithm. Promotions of the average IoU individually utilizing the SASA and CRED module add up to the final promotion of the full model, indicating that the SASA and CRED modules contribute to the different stages of model and data in the training process.

• Journal of Environmental Impact Assessment
• /
• v.17 no.5
• /
• pp.289-297
• /
• 2008

An impervious cover is one of most important factors which effect on a water body environment in a watershed. There are many researches on the impact of an impervious cover on water quality, quantity and ecosystem and most of these researches have been focused on an impervious rate or area in a watershed without considering structure features as like shape, edge, connection of impervious cover. In this study, we focused on a landscape structure which includes shape, density, contiguity, distance, aggregation of land cover type as well as area and rate. The calculation of a landscape indices made to analyse a landscape structure is conducted by applying Fragastats 3.3 program. Osan-cheon watershed where has rapidly urbanized is selected as a study field. Land information for 2002 and 2007 is from land classification maps provided by Ministry of Environment. The result shows that the increasing rate of an impervious cover is more conspicious in Kiheung dam watershed but the fragment of impervious cover areas is shown remarkably in the Osan sub-watershed. The trend of aggregation and connection of impervious covers is increasing. But it was very difficult to say that which type of landscape structure is more beneficial for a watershed management. The implication of this study is to find the need to come over the conventional ways to evaluate landscape structure of a watershed such as rates and areas of impervious cover, and define the importance of landscape feature as like connection, distance, edge density, fragment of impervious covers.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.15 no.1
• /
• pp.133-138
• /
• 2020

This paper introduces a method for improving accuracy using the first convolution layer, which is not used in most modified CNN(: Convolution Neural Networks). In CNN, such as GoogLeNet and DenseNet, the first convolution layer uses only the traditional methods(3×3 convolutional computation, batch normalization, and activation functions), replacing this with RGB-csb. In addition to the results of preceding studies that can improve accuracy by applying RGB values to feature maps, the accuracy is compared with existing CNN using a limited number of images. The method proposed in this paper shows that the smaller the number of images, the greater the learning accuracy deviation, the more unstable, but the higher the accuracy on average compared to the existing CNN. As the number of images increases, the difference in accuracy between the existing CNN and the proposed method decreases, and the proposed method does not seem to have a significant effect.

• Journal of Korean Society of Industrial and Systems Engineering
• /
• v.39 no.3
• /
• pp.83-89
• /
• 2016

The group formation problem of the machine and part is a critical issue in the planning stage of cellular manufacturing systems. The machine-part grouping with alternative process plans means to form machine-part groupings in which a part may be processed not only by a specific process but by many alternative processes. For this problem, this study presents an algorithm based on self organizing neural networks, so called SOM (Self Organizing feature Map). The SOM, a special type of neural networks is an intelligent tool for grouping machines and parts in group formation problem of the machine and part. SOM can learn from complex, multi-dimensional data and transform them into visually decipherable clusters. In the proposed algorithm, output layer in SOM network had been set as one-dimensional structure and the number of output node has been set sufficiently large in order to spread out the input vectors in the order of similarity. In the first stage of the proposed algorithm, SOM has been applied twice to form an initial machine-process group. In the second stage, grouping efficacy is considered to transform the initial machine-process group into a final machine-process group and a final machine-part group. The proposed algorithm was tested on well-known machine-part grouping problems with alternative process plans. The results of this computational study demonstrate the superiority of the proposed algorithm. The proposed algorithm can be easily applied to the group formation problem compared to other meta-heuristic based algorithms. In addition, it can be used to solve large-scale group formation problems.

• Science of Emotion and Sensibility
• /
• v.20 no.2
• /
• pp.161-170
• /
• 2017

Emotion affects many parts of human life such as learning ability, behavior and judgment. It is important to understand human nature. Emotion can only be inferred from facial expressions or gestures, what it actually is. In particular, emotion is difficult to classify not only because individuals feel differently about emotion but also because visually induced emotion does not sustain during whole testing period. To solve the problem, we acquired bio-signals and extracted features from those signals, which offer objective information about emotion stimulus. The emotion pattern classifier was composed of unsupervised learning algorithm with hidden nodes and feature vectors. Restricted Boltzmann machine (RBM) based on probability estimation was used in the unsupervised learning and maps emotion features to transformed dimensions. The emotion was characterized by non-linear classifiers with hidden nodes of a multi layer neural network, named deep belief network (DBN). The accuracy of DBN (about 94 %) was better than that of back-propagation neural network (about 40 %). The DBN showed good performance as the emotion pattern classifier.

• Journal of the Korea Computer Graphics Society
• /
• v.4 no.1
• /
• pp.1-10
• /
• 1998

The idea of level-of-detail based on multiresolution model is gaining popularity as a natural means of handling the complexity regarding the realtime rendering of virtual environments. To generate an effective multiresolution model, we should capture the prominent visual features in the process of simplifying original complex model. In this paper, we incorporate view dependent features such as silhouette features and backface features, to the generation process of multiresolution model. To capture the view directional parameter, we propose multiresolution view sphere. View sphere maps the directional relationship between object surface and the view. Using the view sphere, coherence in the directional space is mapped into spatial coherence in the view sphere. View sphere is generated in multiresolution fashion to simplify the object. To access multiresolution view sphere efficiently, we devise quad tree for the view sphere. We also devise a mechanism for realtime simplification process using proposed view sphere. Using proposed mechanism, regenerating simplified model in realtime is effectively done in the order of number of rendered vertices.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.23 no.3
• /
• pp.313-322
• /
• 2005

Rapid advances in technology and changes in human and cultural activities bring about changes to the earth surface in terms of spatial extension as well as time frame of the changes. Such advances introduce shorter updating frequency of maps and geospatial database. To satisfy these requirements, recent research efforts in the geoinformatics field have been focused on the automation and speeding up of the mapping processes which resulted in products such as the digital photogrammetric workstation, GPSIINS, applications of satellite imagery, automatic feature extraction and the LiDAR system. The possibility of automatically extracting buildings and generating contours from airborne LiDAR data has received much attention because LiDAR data produce promising results. However, compared with the manually derived building footprints using traditional photogrammetric process, more investigation and analysis need to be carried out in terms of accuracy and efficiency. On the other hand, generation of the contours with LiDAR data is more efficient and economical in terms of the quality and accuracy. In this study, the effects of various conditions of the pre-processing phase and the subsequent building extraction and contour generation phases for digital mapping have on the accuracy were investigated.

• Journal of Positioning, Navigation, and Timing
• /
• v.6 no.4
• /
• pp.159-166
• /
• 2017

The position of autonomous driving vehicle is basically acquired through the global positioning system (GPS). However, GPS signals cannot be received in tunnels. Due to this limitation, localization of autonomous driving vehicles can be made through sensors mounted on them. In particular, a 3D Light Detection and Ranging (LIDAR) system is used for longitudinal position error correction. Few feature points and structures that can be used for localization of vehicles are available in tunnels. Since lanes in the road are normally marked by solid line, it cannot be used to recognize a longitudinal position. In addition, only a small number of structures that are separated from the tunnel walls such as sign boards or jet fans are available. Thus, it is necessary to extract usable information from tunnels to recognize a longitudinal position. In this paper, fire hydrants and evacuation guide lights attached at both sides of tunnel walls were used to recognize a longitudinal position. These structures have highly distinctive reflectivity from the surrounding walls, which can be distinguished using LIDAR reflectivity data. Furthermore, reflectivity information of tunnel walls was fused with the road surface reflectivity map to generate a synthetic reflectivity map. When the synthetic reflectivity map was used, localization of vehicles was able through correlation matching with the local maps generated from the current LIDAR data. The experiments were conducted at an expressway including Maseong Tunnel (approximately 1.5 km long). The experiment results showed that the root mean square (RMS) position errors in lateral and longitudinal directions were 0.19 m and 0.35 m, respectively, exhibiting precise localization accuracy.

• Spatial Information Research
• /
• v.6 no.1
• /
• pp.11-23
• /
• 1998

Map generalization functions to visualize the spatial data or to change their scale by changing the level of details of data. Until recently, the studies on map generalization have concentrated more on line features than on point features. However, point features are one of the essential components of digital maps and cannnot be ignored because of the great amount of information they carry. This study, therefore, aimed to find out a detailed procedure of point features' generalization. Particularly, this work chose the distribution pattern of point features as the most important factor in the point generalization in investigating the geometric characteristics of source data. First, it attempted to find out the characteristics of distribution pattern of point features through quadrat analysis with Grieg-Smith method and nearest-neighbour analysis. It then generalized point features through the generalization threshold which did not alter the characteristics of distribution pattern and the removal of redudant point feautres. Therefore, the generalization procedure of point features provided by this work maintained the geometric characteristics as much as possible.