• Proceedings of the Korea Information Processing Society Conference
• /
• 2009.11a
• /
• pp.399-400
• /
• 2009

무마커 기반 증강현실 응용에서 빠르고 정확한 무마커 추적이 수행되어야 한다. 무마커 추적은 등록된 패턴의 특징점들과 입력 영상에서의 특징점들의 매칭을 통하여 수행된다. 매칭에서 이상점은 시차를 크게 유발시키는 요인이 되므로 정확도 향상을 위해서는 이상점을 제거해야 한다. 본 논문에서는 무마커 추적의 정확도 향상을 위한 이상점 제거 방식을 제안한다. 무마커 추적에서 사용되는 SURF 알고리즘을 사용하여 실영상을 캡처하여 실험하였고 정확도 및 실행시간을 비교하였다.

• Journal of Digital Contents Society
• /
• v.16 no.4
• /
• pp.605-613
• /
• 2015

This paper addresses the problem to make sparse the projection matrix in pattern recognition method. Recently, the size of computer program is often restricted in embedded systems. It is very often that developed programs include some constant data. For example, many pattern recognition programs use the projection matrix for dimension reduction. To improve the recognition performance, very high dimensional feature vectors are often extracted. In this case, the projection matrix can be very big. Recently, RSR(roated sparse regression) method[1] was proposed. This method has been proved one of the best algorithm that obtains the sparse matrix. We propose three methods to improve the RSR; outlier removal, sampling and elastic net RSR(E-RSR) in which the penalty term in RSR optimization function is replaced by that of the elastic net regression. The experimental results show that the proposed methods are very effective and improve the sparsity rate dramatically without sacrificing the recognition rate compared to the original RSR method.

• Journal of Korean Society for Geospatial Information Science
• /
• v.19 no.2
• /
• pp.63-73
• /
• 2011

The Pseudorange Corrections (PRC), which are used in DGPS as calibration messages, can contain outliers, noise, and anomalies, and these abnormal events are unpredictable. When those irregular PRC are used, the positioning error gets higher. In this paper, we propose a strategy of detecting and correcting outliers, noise, and anomalies by modeling the changing pattern of PRC through polynomial curve fitting techniques. To validate our strategy, we compared positioning errors obtained without PRC calibation with those with PRC calibration. As a result, we found that our algorithm performs very well; the horizontal RMS error was 3.84 m before the correction and 1.49 m after the correction.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.8 no.5
• /
• pp.1-11
• /
• 2009

TCS Data imply characteristics of traffic conditions. However, there are outliers in TCS data, which can not represent the travel time of the pertinent section, if these outliers are not eliminated, travel time may be distorted owing to these outliers. Various travel time can be distributed under the same section and time because the variation of the travel time is increase as the section distance is increase, which make difficult to calculate the representative of travel time. Accordingly, it is important to grasp travel time characteristics in order to compute the representative of travel time using TCS Data. In this study, after analyzing the variation ratio of the travel time according to the link distance and the level of congestion, the outlier elimination model and the smoothing model for TCS data were proposed. The results show that the proposed model can be utilized for estimating a reliable travel time for a long-distance path in which there are a variation of travel times from the same departure time, the intervals are large and the change in the representative travel time is irregular for a short period.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.22 no.5
• /
• pp.624-630
• /
• 2012

The peer review learning is a method which improves learning outcome of students through feedback between students and the observation and analysis of other students. One of the important problems in a peer review system is to find proper evaluators to each learner considering characteristics of students for improving learning outcomes. Some of peer review systems randomly assign peer review evaluators to learners, or chose evaluators based on limited strategies. However, these systems have a problem that they do not consider various characteristics of learners and evaluators who participate in peer reviews. In this paper, we propose a novel prediction approach of learning outcomes to apply peer review systems considering various characteristics of learners and evaluators. The proposed approach extracts representative attributes from the profiles of students and predicts learning outcomes using various regression models. In order to verify how much outliers affect on the prediction of learning outcomes, we also apply several outlier removal methods to the regression models and compare the predictive performance of learning outcomes. The experiment result says that the SVR model which does not removes outliers shows an error rate of 0.47% on average and has the best predictive performance.

• Journal of Biosystems Engineering
• /
• v.36 no.2
• /
• pp.96-108
• /
• 2011

Several methods of in-field measurements of Nitrogen and other soil properties using cores extracted by a hydraulic soil sampler were evaluated. A prototype core scanner was built to accommodate Veris Technologies commercial Vis-NIRS equipment. The testing result for pH, P and Mg were close to RPD (Ratio of Prediction to Deviation = Standard deviation/RMSE) of 2, however the scanner could not achieve the goal of RPD of 2 on some other properties, especially on nitrate nitrogen ($NO_3$) and potassium (K). In situ NIRS/EC probe showed similar results to the core scanner; pH, P and Mg were close to RPD of 2, while $NO_3$ and K were RPD of 1.5 and 1.2, respectively. Correlations between estimations using the probe and the core scanner were strong, with $r^2$ > 0.7 for P, Mg, Total N, Total C and CEC. Preliminary results for mid-IR spectroscopy showed an $r^2$ of 0.068 and an RMSE for nitrate (N) of 18 ppm, even after the removal of calcareous samples and possible N outlier. After removal of calcareous samples on a larger sample set, results improved considerably with an $r^2$ of 0.64 and RMSE of 6 ppm. However, this was only possible after carbonate samples were detected and eliminated, which would not be feasible under in-field measurements. Testing of $NO_3$ and K ion-selective electrodes (ISEs) revealed promising results, with acceptable errors measuring soil solutions containing nitrate and potassium levels that are typical of production agriculture fields.

• International Journal of Contents
• /
• v.12 no.4
• /
• pp.53-61
• /
• 2016

Although extensively developed, optical music recognition systems have mostly focused on musical symbols (notes, rests, etc.), while disregarding the chord symbols. The process becomes difficult when the images are distorted or slurred, although this can be resolved using optical character recognition systems. Moreover, the appearance of outliers (lyrics, dynamics, etc.) increases the complexity of the chord recognition. Therefore, we propose a new approach addressing these issues. After binarization, un-distortion, and stave and lyric removal of a musical image, a rule-based method is applied to detect the potential regions of chord symbols. Next, a lexicon-driven approach is used to optimally and simultaneously separate and recognize characters. The score that is returned from the recognition process is used to detect the outliers. The effectiveness of our system is demonstrated through impressive accuracy of experimental results on two datasets having a variety of resolutions.

• Korean Journal of Remote Sensing
• /
• v.31 no.4
• /
• pp.293-302
• /
• 2015

Water body extraction is significant for flood disaster monitoring using satellite imagery. Conventional methods have focused on finding an index, which highlights water body and suppresses non-water body such as vegetation or soil area. The Normalized Difference Water Index (NDWI) is typically used to extract water body from satellite images. The drawback of NDWI, however, is that some man-made objects in built-up areas have NDWI values similar to water body. The objective of this paper is to propose a new method that could extract correctly water body with built-up areas in before and after images of flood. We first create a two-element feature vector consisting of NDWI and a Near InfRared band (NIR) and then select a training site on water body area. After computing the mean vector and the covariance matrix of the training site, we classify each pixel into water body based on Mahalanobis distance. We also register before and after images of flood using outlier removal and triangulation-based local transformation. We finally create a change map by combining the before-flooding water body and after-flooding water body. The experimental results show that the overall accuracy and Kappa coefficient of the proposed method were 97.25% and 94.14%, respectively, while those of the NDWI method were 89.5% and 69.6%, respectively.

• Earthquakes and Structures
• /
• v.23 no.6
• /
• pp.549-580
• /
• 2022

This paper assesses the influences of modeling assumptions and uncertainties on the performance of the non-linear finite element (FE) model updating procedure and model clustering method. The results of a shaking table test on a four-story steel moment-resisting frame are employed for both calibrations and clustering of the FE models. In the first part, simple to detailed non-linear FE models of the test frame is calibrated to minimize the difference between the various data features of the models and the structure. To investigate the effect of the specified data feature, four of which include the acceleration, displacement, hysteretic energy, and instantaneous features of responses, have been considered. In the last part of the work, a model-based clustering approach to group models of a four-story frame with similar behavior is introduced to detect abnormal ones. The approach is a composition of property derivation, outlier removal based on k-Nearest neighbors, and a K-means clustering approach using specified data features. The clustering results showed correlations among similar models. Moreover, it also helped to detect the best strategy for modeling different structural components.

• Journal of Korean Society of Transportation
• /
• v.22 no.7 s.78
• /
• pp.7-16
• /
• 2004

The homogeneous road section is defined as one consisted of similar traffic characteristics focused on demand and supply. The criteria, in the aspect of demand, are the diverging rate and the ratio of green time to cycle time at signalized intersection, and distance between the signalized intersections. The criteria, in that or supply, are the traffic patterns such as traffic volume and its speed. In this study, the effective method to generate valuable data, pointing out the problems of removal method of obscure data, is proposed using data collected from Gonjiam IC to Jangji IC on the national highway No.3. Travel times are collected with licence matching method and traffic volume and speed are collected from detectors. Futhermore, the method of selecting homogeneous road section is proposed considering demand and supply aspect simultaneously. This method using outlier filtering algorithm can be applied to generate the travel time forecasting model and to revise the obscured of missing data transmitting from detectors. The point and link data collected at the same time on the rational highway can be used as a basis predicting the travel time and revising the obscured data in the future.