http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
모바일 네트워크를 이용한 상호정보량 기반의 다수 이동 물체 추적 알고리즘 설계
임진홍(Jinhong Lim),김현진(H.Jin Kim) 제어로봇시스템학회 2017 제어·로봇·시스템학회 논문지 Vol.23 No.3
Tracking multiple moving targets involves many issues such as the sensors’ limited field of view, and the unknown number of targets with unknown dynamics. This paper performs multi-target tracking and target number estimation using a Gaussian mixture probability hypothesis density (GM-PHD) filter. Mutual information is calculated by approximate computation in nonparametric methods and the network of sensing robots is controlled to detect the maximum number of targets by maximizing the mutual information. In addition, we propose the motion pattern learning method using multiple Gaussian Process (GP) models to enhance the multi-target tracking performance for various types of movement by accurately predicting future target states. Among the multiple motion patterns learned in advance, the most proper pattern is assigned by the maximum likelihood principle. The performance of the proposed algorithm is validated via simulation in terms of the accuracy of target number estimation, and the reliability of multi-target tracking.
다양한 센서 기반의 침입체 탐지, 분류 및 추적 알고리즘 개발
김원철(Wonchul Kim),임진홍(Jinhong Lim),김태완(Taewan Kim),손영동(Youngdong Son),김현진(H. Jin Kim),김진영(Jinyoung Kim),홍수연(Sooyoun Hong),김한동(Handong Kim) 제어로봇시스템학회 2017 제어·로봇·시스템학회 논문지 Vol.23 No.11
Surveillance is one of the major applications in wireless sensor network areas, and it is important to detect, classify and localize the targets. In this paper, we divide the research into two sections: (1) detecting and classifying the targets and (2) localizing them. To detect and classify multiple moving targets, we use acoustic and seismic sensors, and we analyze raw data from the sensors in both time and frequency domains. In this process, we must decide which features are useful for the classification to improve the performance and make it work in real time. Thus, we exploit Weibull likelihood and short-time Fourier transform (STFT) to extract the features as a sampling method. Then, we implement a support vector machine (SVM) and a neural network to classify the type of targets based on those features. Using the suggested algorithms, the proposed classifiers provide more accurate performance than the method that analyzes the raw data from only the frequency or time domain. For localization, Gaussian Process Regression (GPR) is used to estimate the relative location that corresponds to the received signal strength indication (RSSI) data. We also demonstrate the simultaneous localization with the process of detection and classification in real time. Finally, experimental results validate the suggested algorithm.