• Proceedings of the Korea Information Processing Society Conference
• /
• 2016.04a
• /
• pp.633-635
• /
• 2016

Convolutional Networks (ConvNets) are powerful models that learn hierarchies of visual features, which could also be used to obtain image representations for transfer learning. The basic pipeline for transfer learning is to first train a ConvNet on a large dataset (source task) and then use feed-forward units activation of the trained ConvNet as image representation for smaller datasets (target task). Our key contribution is to demonstrate superior performance of multiple ConvNet layer features over single ConvNet layer features. Combining multiple ConvNet layer features will result in more complex feature space with some features being repetitive. This requires some form of feature selection. We use AdaBoost with single stumps to implicitly select only distinct features that are useful towards classification from concatenated ConvNet features. Experimental results show that using multiple ConvNet layer activation features instead of single ConvNet layer features consistently will produce superior performance. Improvements becomes significant as we increase the distance between source task and the target task.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.12 no.1
• /
• pp.35-41
• /
• 2017

We propose a face recognition and notification system that can transform visual face information into tactile signals in order to help visually impaired people. The proposed system consists of a glasses type camera, a mobile computer and an electronic cane. The glasses type camera captures the frontal view of the user, and sends this image to mobile computer. The mobile computer starts to search for human's face in the image when obstacles are detected by ultrasonic sensors. In a case that human's face is detected, the mobile computer identifies detected face. At this time, Adaboost and compressive sensing are used as a detector and a classifier, respectively. After the identification procedures of the detected face, the identified face information is sent to controller attached to a cane using a Bluetooth communication. The controller generates motor control signals using Pulse Width Modulation (PWM) according to the recognized face labels. The vibration motor generates vibration patterns to inform the visually impaired person of the face recognition result. The experimental results of face recognition and notification system show that proposed system is helpful for visually impaired people by providing person identification results in front of him/her.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2015.07a
• /
• pp.254-257
• /
• 2015

차량 인식 기술은 운전자에게 차량 충돌과 같은 위험요소를 사전에 인식시키거나 차량을 자동으로 제어하는 기술로 각광 받고 있다. 본 논문에서는 입력 영상에서 차량이 나타날 수 있는 관심 영역을 설정한 다음 미리 학습된 검출기를 통한 Haar-like와 Adaboost 알고리즘으로 차량 후보 영역을 검출하고 중복된 영역을 제거하기 위인식 기술해 클러스터링 기법을 적용하고, 칼만필터로 프레임 영상에서 차량을 추적 하고, 다시 중복된 영역에 대해 클러스터링 기법을 적용하는 방법을 제안하였다.

• Journal of the Korean Data and Information Science Society
• /
• v.25 no.6
• /
• pp.1283-1291
• /
• 2014

In this paper, we propose forward stagewise algorithm when data are very large or coming in batches sequentially over time. In this situation, ordinary boosting algorithm for large scale data and data batch stream may be greedy and have worse performance with class noise situations. To overcome those and apply to large scale data or data batch stream, we modify the forward stagewise algorithm. This algorithm has better results for both large scale data and data batch stream with or without concept drift on simulated data and real data sets than boosting algorithms.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2013.11a
• /
• pp.120-123
• /
• 2013

본 논문에서는 키넥트 센서의 RGB영상을 이용하여 얼굴을 검출하고 검출된 영역의 깊이정보를 템플릿으로 사용하여 다수개의 얼굴을 추적하는 방법을 제안한다. 이 논문은 [1]의 단일 얼굴 추적방법을 다수의 얼굴을 추적하도록 확장한 것이다. 다수의 얼굴추적을 실시간으로 처리하기 위하여 영상을 down sampling 하여 사용한다. 얼굴 검출은 기본적으로 기존의 Adaboost 방법을 사용하나, 피부색만을 이용, 탐색영역을 최대한 축소하여 수행 시간 및 오검출율을 줄인다. 얼굴추적은 깊이정보를 템플릿으로 하며, 깊이값에 따라 크기, 탐색영역을 조정하고, 또한 일정 프레임마다 얼굴을 검출하며 겹침, 새로 나타남, 영상 밖으로 사라짐 등의 얼굴추적 시 발생하는 문제를 해결한다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2010.06c
• /
• pp.490-495
• /
• 2010

본 연구에서는 효과적인 손 제스처 인식을 위하여 다중 특징을 이용한 견고한 손 추적 방법을 제시한다. 기존의 많은 손추적 장치들이 칼라 정보나 모션 정보와 같은 단일한 정보를 바탕으로 손을 검출하고, 이를 바탕으로 손의 추적하는 방법들을 제시하고 있다. 이러한 방법들의 경우에는 손 추적 중에 환경이나 상황이 변하게 되면, 손추적의 정확도가 현저하게 떨어지게 된다. 본 연구에서는 이러한 문제점들을 보완하기 위하여, Adaboost를 이용한 손 검출, 역투영을 기반으로 손 색상을 이용한 추적, KLT를 바탕으로 한 모션 추적을 이용한 검출을 동시에 수행하며, 각 센서의 추적 결과에 대한 칼만 필터 적용뿐 아니라, 각 센서 정보를 통합하여 견고한 결과를 얻기 위한 방법을 제시한다. 이를 바탕으로 손제스처 인식 시스템을 개발하였으며, 개발된 제스처 인식을 바탕으로 비디오 플레이를 제어하는 시스템을 구현하였다.

• Journal of the Institute of Convergence Signal Processing
• /
• v.13 no.2
• /
• pp.77-82
• /
• 2012

This paper presents a hardware-implemented face regions stabilization algorithm that stabilizes facial regions using the locations and sizes of human faces found by a face detection system. Face detection algorithms extract facial features or patterns determining the presence of a face from a video source and detect faces via a classifier trained on example faces. But face detection results has big variations in the detected locations and sizes of faces by slight shaking. To address this problem, the high frequency reduce filter that reduces variations in the detected face regions by taking into account the face range information between the current and previous video frames are implemented in addition to center distance comparison and zooming operations.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.12 no.5
• /
• pp.843-848
• /
• 2008

Most of face detection technique uses information from the face of the movement. The traditional face detection method is to use difference picture method ate used to detect movement. However, most do not consider this mathematical approach using real-time or real-time implementation of the algorithm is complicated, not easy. This paper, the first to detect real-time facial image is converted YCbCr and RGB video input. Next, you convert the difference between video images of two adjacent to obtain and then to conduct Glassfire Labeling. Labeling value compared to the threshold behavior Area recognizes and converts video extracts. Actions to convert video to conduct face detection, and detection of facial characteristics required for the extraction and use of AdaBoost algorithm.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2009.05a
• /
• pp.403-406
• /
• 2009

In this paper, we propose a method for deciding permission from the ATM(Automated Teller Machine) using face detection. First, we extract skin areas and make candidate face images from an input image, and then detect a face using Adaboost(Adaptive Boosting) algorithm. Next, proposed method executes a template matching for making a decision on whether to wear accessories like sunglasses or a mask in detected face image. Finally, this method decides whether to permit ATM service using this result. Experimental results show that proposed method performed well at indoors ATM environment for detecting whether to wear accessories.

• Computers and Concrete
• /
• v.32 no.3
• /
• pp.233-246
• /
• 2023

Accurate prediction of concrete compressive strength can minimize the need for extensive, time-consuming, and costly mixture optimization testing and analysis. This study attempts to enhance the prediction accuracy of compressive strength using stacking ensemble machine learning (ML) with feature engineering techniques. Seven alternative ML models of increasing complexity were implemented and compared, including linear regression, SVM, decision tree, multiple layer perceptron, random forest, Xgboost and Adaboost. To further improve the prediction accuracy, a ML pipeline was proposed in which the feature engineering technique was implemented, and a two-layer stacked model was developed. The k-fold cross-validation approach was employed to optimize model parameters and train the stacked model. The stacked model showed superior performance in predicting concrete compressive strength with a correlation of determination (R²) of 0.985. Feature (i.e., variable) importance was determined to demonstrate how useful the synthetic features are in prediction and provide better interpretability of the data and the model. The methodology in this study promotes a more thorough assessment of alternative ML algorithms and rather than focusing on any single ML model type for concrete compressive strength prediction.