• ETRI Journal
• /
• 제45권2호
• /
• pp.329-337
• /
• 2023

For safe last-mile autonomous robot delivery services in complex environments, rapid and accurate collision prediction and detection is vital. This study proposes a suitable neural network model that relies on multiple navigation sensors. A light detection and ranging technique is used to measure the relative distances to potential collision obstacles along the robot's path of motion, and an accelerometer is used to detect impacts. The proposed method tightly couples relative distance and acceleration time-series data in a complementary fashion to minimize errors. A long short-term memory, fully connected layer, and SoftMax function are integrated to train and classify the rapidly changing collision countermeasure state during robot motion. Simulation results show that the proposed method effectively performs collision prediction and detection for various obstacles.

• 한국멀티미디어학회논문지
• /
• 제25권4호
• /
• pp.575-582
• /
• 2022

In the manufacturing environment, research to minimize robot collisions with human beings have been widespread, but in order to interact with robots, it is important to precisely recognize and predict human actions. In this research, after enhancing performance by applying group normalization to the Hourglass model to detect the operator motion, the skeleton was estimated and data were created using this model. And then, three types of operator's movements were recognized using LSTM. As results of the experiment, the accuracy was enhanced by 1% using group normalization, and the recognition accuracy was 99.6%.

• 제어로봇시스템학회논문지
• /
• 제10권9호
• /
• pp.779-784
• /
• 2004

This paper describes a sonic polygonal multiple reflection range sensor (SPMRS), which uses multiple reflection properties usually ignored in ultrasonic sensors as disturbances or noises. Targets such as a plane, corner, edge, or cylinder in indoor environments can easily be detected by the multiple reflection patterns obtained with a SPMRS system. Target classification and feature data extraction, such as distance and azimuth to the target, are computed simultaneously by considering the geometrical relationships between the detected targets, and finally the environment model is generated by refining the detected targets. In addition, the narrow field of view of a sonar range sensor is increased and the scanning time is reduced by active motion of the SPMRS stepping servomechanism.

• 제어로봇시스템학회논문지
• /
• 제10권9호
• /
• pp.755-762
• /
• 2004

In this research, EOG(Electrooculogram) signal was analyzed to predict the subject's intention using a fuzzy classifier. The fuzzy classifier is built automatically using the EOG data and evolutionary algorithms. An assistant robot manipulator in redundant configuration has been developed, which operates according to the EOG signal classification results. For automatic fuzzy model construction without any experts' knowledge, an evolutionary algorithm with the new representation scheme, design of adequate fitness function and evolutionary operators, is proposed. The proposed evolutionary algorithm can optimize the number of fuzzy rules, the number of fuzzy membership functions, parameter values for the each membership functions, and parameter values for the consequent parts. It is shown that the fuzzy classifier built by the proposed algorithm can classify the EOG data efficiently. Intelligent motion planner that consists of several neural networks are used for control of robot manipulator based upon EOG classification results.

• 대한인간공학회지
• /
• 제10권1호
• /
• pp.3-10
• /
• 1991

The teach pendant and keyboard have been used as an input device of control command in human-robot sustem. But, many problems occur in case that the usef is a novice. So, speech recognition system is required to communicate between a human and the robot. In this study, Korean voice commands, eitht robot commands, and ten digits based on the broad phonetic analysis are described. Applying broad phonetic analysis, phonemes of voice commands are divided into phoneme groups, such as plosive, fricative, affricative, nasal, and glide sound, having similar features. And then, the feature parameters and their ranges to detect phoneme groups are found by minimax method. Classification rules are consisted of combination of the feature parameters, such as zero corssing rate(ZCR), log engery(LE), up and down(UD), formant frequency, and their ranges. Voice commands were recognized by the classification rules. The recognition rate was over 90 percent in this experiment. Also, this experiment showed that the recognition rate about digits was better than that about robot commands.

• 한국정보통신학회:학술대회논문집
• /
• 한국정보통신학회 2018년도 추계학술대회
• /
• pp.477-479
• /
• 2018

본 연구는 인공신경망의 한 종류인 LSTM(Long Short-Term Memory) 기법을 활용하기 위하여 진행하였다. UTD(University of Texas at Dallas)가 공개한 27종 동작 데이터 중 3축 가속도 및 각속도 데이터를 기본 LSTM 및 Deep Residual Bidir-LSTM 기법에 적용하여 행동을 분류해 보았다.

• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2000년도 추계학술발표논문집 (하)
• /
• pp.931-934
• /
• 2000

움직임 추정 및 보상을 위한 방법 중 가장 많이 사용하는 블록 정합 방법은 어떤 평가 함수와 탐색방법(Search Procedure)을 사용했느냐에 따라 그 성능이 달라지게 된다. 본 논문에서는 평가 함수로써 평균 제곱 오차(Mean Squared Error; MSE), 평균 절대값 오차(Mean Absolute Error; MAE), 화소 차분류(Pel Difference Classification: PDC)을, 탐색 방법으로써 전체 탐색 방법(Full Search Method : FSM), 3단계 탐색 방법(Three Step Search : TSS), 대각 탐색 방법(Cross Search Algorithm ;CSA)을 사용하여 이들의 성능을 각각 비교 분석하여 봄으로써 블록 정합 방법을 이용한 움직임 추정에 대한 전반적인 이해를 도모하고자 한다.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 2003년도 학술회의 논문집 정보 및 제어부문 B
• /
• pp.830-835
• /
• 2003

This paper describes a sonic polygonal multiple reflection range sensor (SPMRS), which uses multiple reflection properties usually ignored in ultrasonic sensors as disturbances or noises. Targets such as a plane, corner, edge, or cylinder in indoor environments can easily be detected by the multiple reflection patterns obtained with a SPMRS system. Target classification and feature data extraction, such as distance and azimuth to the target, are computed simultaneously by considering the geometrical relationships between the detected targets, and finally the environment model is generated by refining the detected targets. In addition, the narrow field of view of a sonar range sensor is increased and the scanning time is reduced by active motion of the SPMRS stepping servomechanism.

• 인터넷정보학회논문지
• /
• 제21권6호
• /
• pp.81-94
• /
• 2020

This study developed evaluation indicators for the comfort experience of virtual reality (VR) headsets by classifying, defining, and weighting cybersickness-causing factors using the Delphi research method and analytic hierarchical process (AHP) approach. Four surveys were conducted with 20 experts on VR motion sickness. The expert surveys involved the 1) classification and definition of cybersickness-causing dimensions, classification of sub-factors for each dimension, and selection of evaluation indicators, 2) self-reassessment of the results of each step, 3) validity revaluation, and 4) final weighting calculation. Based on the surveys, the evaluation indicators for the comfort experience of VR headsets were classified into eight sub-factors: field of view (FoV)-device FoV, latency-device latency, framerate-device framerate, V-sync-device V-sync, rig-camera angle view, rig-no-parallax point, resolution-device resolution, and resolution-pixels per inch (PPI). A total of six dimensions and eight sub-factors were identified; sub-factor-based evaluation indicators were also developed.

• ETRI Journal
• /
• 제44권2호
• /
• pp.327-338
• /
• 2022

Classification models for human action recognition require robust features and large training sets for good generalization. However, data augmentation methods are employed for imbalanced training sets to achieve higher accuracy. These samples generated using data augmentation only reflect existing samples within the training set, their feature representations are less diverse and hence, contribute to less precise classification. This paper presents new data augmentation and action representation approaches to grow training sets. The proposed approach is based on two fundamental concepts: virtual video generation for augmentation and representation of the action videos through robust features. Virtual videos are generated from the motion history templates of action videos, which are convolved using a convolutional neural network, to generate deep features. Furthermore, by observing an objective function of the genetic algorithm, the spatiotemporal features of different samples are combined, to generate the representations of the virtual videos and then classified through an extreme learning machine classifier on MuHAVi-Uncut, iXMAS, and IAVID-1 datasets.