• KIISE Transactions on Computing Practices
• /
• v.23 no.2
• /
• pp.104-109
• /
• 2017

We propose a new machine learning algorithm which analyzes daily activity patterns of users from multi-modal wearable sensor data. The proposed model learns and extracts activity patterns using input from wearable devices in real-time. Inspired by cue integration of human's property, we constructed gated multi-modal neural networks which integrate wearable sensor input data selectively by using gate modules. For the experiments, sensory data were collected by using multiple wearable devices in restaurant situations. As an experimental result, we first show that the proposed model performs well in terms of prediction accuracy. Then, the possibility to construct a knowledge schema automatically by analyzing the activation patterns in the middle layer of our proposed model is explained.

• The KIPS Transactions:PartC
• /
• v.11C no.3
• /
• pp.301-308
• /
• 2004

Misuse IDS is known to have an acceptable accuracy but suffers from high rates of false alarms. We show a behavior based alarm reduction with a memory-based machine learning technique. Our extended form of IBL, (XIBL) examines SNORT alarm signals if that signal is worthy sending signals to security manager. An experiment shows that there exists an apparent difference between true alarms and false alarms with respect to XIBL behavior This gives clear evidence that although an attack in the network consists of a sequence of packets, decisions over Individual packet can be used in conjunction with misuse IDS for better performance.

• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.5
• /
• pp.187-194
• /
• 2021

A person's behavioral recognition is the recognition of what a person does according to joint movements. To this end, we utilize computer vision tasks that are utilized in image processing. Human behavior recognition is a safety accident response service that combines deep learning and CCTV, and can be applied within the safety management site. Existing studies are relatively lacking in behavioral recognition studies through human joint keypoint extraction by utilizing deep learning. There were also problems that were difficult to manage workers continuously and systematically at safety management sites. In this paper, to address these problems, we propose a method to recognize risk behavior using only joint keypoints and joint motion information. AlphaPose, one of the pose estimation methods, was used to extract joint keypoints in the body part. The extracted joint keypoints were sequentially entered into the Long Short-Term Memory (LSTM) model to be learned with continuous data. After checking the behavioral recognition accuracy, it was confirmed that the accuracy of the "Lying Down" behavioral recognition results was high.

• Journal of the Korean Academy of Child and Adolescent Psychiatry
• /
• v.5 no.1
• /
• pp.93-101
• /
• 1994

This study was performed to assess the amount of altered serotonergic responsivity in individuals with repeated aggressive behaviors compared with normal controls. Sixteen aggressive(delinquent criminals with repeated aggressive behaviors) and seventeen controls(medical college students) were selected and assessed their severity of aggression by several psychological instruments. The platelet $^3H$-imipramine binding sites which is known to correlate the serotonergic function of the central nervous system were measured. The results are as follows. 1) Mean scores of physical aggression in the aggressive subjects were found to be significantly higher than normal controls(p<.01). And impulsivity, hostility, psychoticism in the aggressives were found higher than controls, also. 2) In the paltelet $^3H$-imipramine binding, the aggressives had a tendency of reduced maximal binding sites(Bmax) comparing with controls(p=.0841). 3) There was no statistically significant differences between two groups in the binding coefficients(Kd) of platelet $^3H$ Imipramine binding. 4) The value of maximal binding sites(Bmax) showed significant inverse correlations with aggressive scale scores of PFAV(r=-.6311), and physical aggression scale scores of CTS(r=-.5377).

• Journal of the Institute of Electronics Engineers of Korea SC
• /
• v.46 no.4
• /
• pp.7-16
• /
• 2009

A robot must be able to generate various skills to achieve given tasks intelligently and reasonably. The robot must first learn affordances to generate the skills. An affordance is defined as qualities of objects or environments that induce actions. Affordances can be usefully used to generate skills. Most tasks require sequential and goal-oriented behaviors. However, it is usually difficult to accomplish such tasks with affordances alone. To accomplish such tasks, a skill is constructed with an affordance and a soft behavioral motivation switch for reflecting goal-oriented elements. A skill calculates a behavioral motivation as a combination of both presently perceived information and goal-oriented elements. Here, a behavioral motivation is the internal condition that activates a goal-oriented behavior. In addition, a robot must be able to execute sequential behaviors. We construct skill networks by using generated skills that make action selection feasible to accomplish a task. A robot can select sequential and a goal-oriented behaviors using the skill network. For this, we will first propose a method for modeling and learning Bayesian networks that are used to generate affordances. To select sequential and goal-oriented behaviors, we construct skills using affordances and soft behavioral motivation switches. We also propose a method to generate the skill networks using the skills to execute given tasks. Finally, we will propose action-selection-mechanism to select sequential and goal-oriented behaviors using the skill network. To demonstrate the validity of our proposed methods, "Searching-for-a-target-object", "Approaching-a-target-object", "Sniffing-a-target-object", and "Kicking-a-target-object" affordances have been learned with GENIBO (pet robot) based on the human teaching method. Some experiments have also been performed with GENIBO using the skills and the skill networks.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2005.11a
• /
• pp.305-308
• /
• 2005

시각 센서를 이용한 환경 및 상황 인식은 로봇의 자동화된 행동을 위해서 매우 중요하다. 실제 환경에서 사람은 주위를 인식할 때 여러 단계의 인식과정을 거친다. 효율적이고 정확한 환경 인식을 위해서는 지능형 로봇의 인식 또한 사람의 인식과정과 같이 다단계로 이루어져야 한다. 또한 실제 환경은 유동적이며 많은 불확실성을 가지고 있으므로 불확실한 상황에 강인한 인식 방법이 필요하다. 이러한 불확실성을 내포한 환경 및 상황 인식에는 베이지안 네트워크를 이용한 인식이 강인하나 복잡한 환경을 하나의 베이지안 네트워크로 인식하는 것은 어렵다. 이 논문에서는 복잡하고 불확실한 환경 인식을 위한 여러 베이지안 네트워크를 사람의 인식과 같은 다단계의 인식 과정으로 구성된 행동 네트워크 기반으로 결합하는 앙상블 기법을 제안한다. 불확실한 상황을 적용한 환경 실험과 로봇 시뮬레이터를 이용한 로봇 실험으로 베이지안 네트워크 앙상블 기법이 환경 인식에 효과적인 것을 확인할 수 있었다.

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.51 no.4
• /
• pp.160-166
• /
• 2002

Since conventional mobile robot control with one module has limitation to solve complex problems, there have been a variety of works on combining multiple modules for solving them. Recently, many researchers attempt to develop mobile robot controllers using artificial life techniques. In this paper, we develop a mobile robot controller using cellular automata based neural networks, where complex tasks are divided to simple sub-tasks and optimal neural structure of each sub-task is explored by genetic algorithm. Neural network modules are combined dynamically using the action selection mechanism, where basic behavior modules compete each other by inhibition and cooperation. Khepera mobile robot simulator is used to verify the proposed model. Experimental results show that complex behaviors emerge from the combination of low-level behavior modules.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2021.01a
• /
• pp.213-216
• /
• 2021

본 논문에서는 Text Classification에 사용된 딥러닝 모델을 적용하여 행동 인식, 손동작 인식 및 감정 인식 방법을 제안한다. 먼저 라이브러리를 사용하여 영상에서 특징 추출 후 식을 적용하여 특징의 벡터를 저장한다. 이를 Conv1D, Transformer, GRU를 결합한 모델에 학습시킨다. 이 방법을 통해 하나의 딥러닝 모델을 사용하여 다양한 분야에 적용할 수 있다. 제안한 방법을 사용해 SYSU 3D HOI 데이터셋에서 99.66%, eNTERFACE' 05 데이터셋에 대해 99.0%, DHG-14 데이터셋에 대해 95.48%의 클래스 분류 정확도를 얻을 수 있었다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.05a
• /
• pp.633-635
• /
• 2023

센서 데이터를 예측 또는 분석하여 시스템을 제어하거나 모니터링할 수 있다. 센서 데이터를 이용한 예측의 신뢰성을 확보하기 위해서는 데이터의 적절한 빈도수가 중요하다. 이를 위해 본 논문에서는 Diffusion Model을 사용한 센서 데이터 주파수 보간을 통해 행동을 예측하는 방법을 제시하고자 한다. 주파수 보간은 반려동물 행동별 25hz 센서 데이터로 학습된 Diffusion Model을 사용한다. 학습된 Diffusion Model에 1hz 센서 데이터와 가우시안 노이즈를 결합한 데이터를 입력으로 사용해 센서데이터를 보간한다. 제안한 방법은 CNN-LSTM 모델 학습 후 예측 성능 비교를 통해 검증한다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2007.11a
• /
• pp.63-65
• /
• 2007

본 연구에서는 비디오 영상데이터로부터 인간의 행동패턴의 인식기술 및 상황인식 기법을 소개하고 이를 활용한 실용적 응용으로서 지능형 감시시스템을 제안한다. 순차적 영상신호에서 형태기반의 정적 특징과 목표물의 움직임 요소를 측정한 동적 특징을 결합한 형태의 특징 표현 및 추출기법과 행동패턴 및 상황패턴에 대한 인식 모델을 제시하고 구현한다. 모듈구조의 시스템에서 영상처리 모듈과 패턴인식 모듈은 특징추출 및 인식과정을 수행하며, 감시영상에 대한 상황판단 기능은 데이터베이스 모듈과 연동하여 효과적인 검색기능과 경보기능 등을 지원한다. 이러한 기능은 기존의 시스템에서 운영자의 지속적인 감시작업과 상황판단 작업을 보조 또는 대행하여 수행할 수 있을 뿐만 아니라 데이터저장 공간을 획기적으로 줄이고 부수적으로 효율적인 영상의 조회기능 및 추적기능 등의 유용한 인터페이스를 지원한다.