• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.11
• /
• pp.491-500
• /
• 2021

In this study, we propose a method for detecting fall behavior using MS Kinect v2 RGBD Camera-based Human-Skeleton Keypoints and a 2-Stacked Bi-LSTM model. In previous studies, skeletal information was extracted from RGB images using a deep learning model such as OpenPose, and then recognition was performed using a recurrent neural network model such as LSTM and GRU. The proposed method receives skeletal information directly from the camera, extracts 2 time-series features of acceleration and distance, and then recognizes the fall behavior using the 2-Stacked Bi-LSTM model. The central joint was obtained for the major skeletons such as the shoulder, spine, and pelvis, and the movement acceleration and distance from the floor were proposed as features of the central joint. The extracted features were compared with models such as Stacked LSTM and Bi-LSTM, and improved detection performance compared to existing studies such as GRU and LSTM was demonstrated through experiments.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.22 no.2
• /
• pp.127-133
• /
• 2021

A recent study of people physically falling focused on analyzing the motions of the falls using a recurrent neural network (RNN) and a deep learning approach to get good results from detecting 2D human poses from a single color image. In this paper, we investigate a detection method for estimating the position of the head and shoulder keypoints and the acceleration of positional change using the skeletal keypoints information extracted using PoseNet from an image obtained with a low-cost 2D RGB camera, increasing the accuracy of judgments about the falls. In particular, we propose a fall detection method based on the characteristics of post-fall posture in the fall motion-analysis method. A public data set was used to extract human skeletal features, and as a result of an experiment to find a feature extraction method that can achieve high classification accuracy, the proposed method showed a 99.8% success rate in detecting falls more effectively than a conventional, primitive skeletal data-use method.

• Journal of Sensor Science and Technology
• /
• v.24 no.1
• /
• pp.47-53
• /
• 2015

Fall-related injuries in elderly people are a major health care problem. This paper introduces determination of fall direction before impact using support vector machine (SVM). Once a falling phase is detected, dynamic characteristic parameters measured by the accelerometer and gyroscope and then processed by a Kalman filter are used in the SVM to determine the fall directions, i.e., forward (F), backward (B), rightward (R), and leftward (L). This paper compares the determination sensitivities according to the selected parameters for the SVM (velocities, tilt angles, vs. accelerations) and sensor attachment locations (waist vs. chest) with regards to the binary classification (i.e., F vs. B and R vs. L) and the multi-class classification (i.e., F, B, R, vs. L). Based on the velocity of waist which was superior to other parameters, the SVM in the binary case achieved 100% sensitivities for both F vs. B and R vs. L, while the SVM in the multi-class case achieved the sensitivities of F 93.8%, B 91.3%, R 62.3%, and L 63.6%.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2022.10a
• /
• pp.123-125
• /
• 2022

In this paper, we introduce a system applied to long short-term memory using Yolo-pose. Using Yolo-pose from image data, data divided into daily life and falls are extracted and applied to LSTM for learning. In order to prevent overfitting, training is performed 8 to 2 validation and is represented by a confusion matrix. The result of Yolo-pose recorded 100% of both sensitivity and specificity, confirming that daily life and falls were well distinguished.

• Journal of the Korean Society of Marine Environment & Safety
• /
• v.26 no.6
• /
• pp.601-605
• /
• 2020

In large ships with complex structures, it is difficult to locate workers. In particular, it is not easy to detect when a worker falls down, making it difficult to respond quickly. Thus, research is being conducted to detect fallen workers using a camera or by attaching a device to the body. Existing image-based fall detection systems have been designed to detect a person's body parts; hence, it is difficult to detect them in various ships and postures. In this study, the entire fall area was extracted and deep learning was used to detect the fallen shipworker based on the image. The data necessary for learning were obtained by recording falling states at the shipyard. The amount of learning data was augmented by flipping, resizing, and rotating the image. Performance evaluation was conducted with precision, reproducibility, accuracy, and a low error rate. The larger the amount of data, the better the precision. In the future, reinforcing various data is expected to improve the effectiveness of camera-based fall detection models, and thus improve safety.

• Pediatric Infection and Vaccine
• /
• v.21 no.1
• /
• pp.22-28
• /
• 2014

Purpose: This study aimed at determining the detection rate of respiratory viruses and at investigating the risk factors associated with respiratory virus detection in young infants. Methods: From September 2011 to August 2012, nasopharyngeal swabs were obtained from 227 infants aged ${\leq}90$ days with suspected infectious diseases, including sepsis. We performed a retrospective analysis of their clinical characteristics. The prevalence of respiratory viruses in their nasopharyngeal swabs was assayed by real-time polymerase chain reaction (real-time PCR). Results: In total, 157 (69.2%) infants had more than one of the following respiratory viruses: respiratory syncytial virus (n=75), rhinovirus (n=42), influenza virus (n=18), parainfluenza virus (n=15), human metapneumovirus (n=9), coronavirus (n=9), adenovirus (n=4), and bocavirus (n=3). During the same period, bacterial infections were confirmed in 24 infants (10.6%). The detection of respiratory viruses was significantly associated with the presence of cough, a family history of respiratory illness, and a seasonal preference (fall/winter). Using logistic regression analysis, these 3 variables were also identified as significant risk factors. During fall and winter, detection of respiratory viruses was significantly higher in infants who did not have a bacterial infection. Conclusion: Respiratory virus is an important pathogen in young infants admitted to a hospital, who are suspected with infectious diseases. Detection of respiratory viruses in young infants was associated with seasonality (fall/winter), presence of respiratory symptoms and a family history of respiratory illness.

• Journal of Internet Computing and Services
• /
• v.13 no.4
• /
• pp.23-31
• /
• 2012

Nowadays, recognizing human activities is an important subject; it is exploited widely and applied to many fields in real-life, especially in health care and context aware application. Research achievements are mainly focused on activities of daily living which are useful for suggesting advises to health care applications. Falling event is one of the biggest risks to the health and well-being of the elderly especially in independent living because falling accidents may be caused from heart attack. Recognizing this activity still remains in difficult research area. Many systems equipped wearable sensors have been proposed but they are not useful if users forget to wear the clothes or lack ability to adapt themselves to mobile systems without specific wearable sensors. In this paper, we develop a novel method based on analyzing the change of acceleration, orientation when the fall occurs and measure their similarity to featured fall patterns. In this study, we recruit five volunteers in our experiment including various fall categories. The results are effective for recognizing fall activity. Our system is implemented on G1 smart phone which are already plugged accelerometer and orientation sensors. The popular phone is used to get data from accelerometer and results showthe feasibility of our method and significant contribution to fall detection.

• ETRI Journal
• /
• v.44 no.4
• /
• pp.654-671
• /
• 2022

Accidental falls frequently occur during activities of daily living. Although many studies have proposed various accident detection methods, no high-performance accident detection system is available. In this study, we propose a method for integrating data and accident detection algorithms presented in existing studies, collect new data (from two stunt performers and 15 people over age 60) using a developed wearable device, demonstrate new features and related accident detection algorithms, and analyze the performance of the proposed method against existing methods. Comparative analysis results show that the newly defined features extracted reflect more important risk factors than those used in existing studies. Further, although the traditional algorithms applied to integrated data achieved an accuracy (AC) of 79.5% and a false positive rate (FPR) of 19.4%, the proposed accident detection algorithms achieved 97.8% AC and 2.9% FPR. The high AC and low FPR for accidental falls indicate that the proposed method exhibits a considerable advancement toward developing a commercial accident detection system.

• Proceedings of the IEEK Conference
• /
• 1999.11a
• /
• pp.1057-1060
• /
• 1999

For efficient storage and retrieval of large video data sets, automatic video scene change detection is a necessary tool. Video scene changes fall into two categories, namely fast and gradual scene changes. The gradual scene change effects include, dissolves, wipes, fades, etc. Although currently existing algorithms are able to detect fast scene changes quite accurately, the detection of gradual scene changes continue to remain a difficult problem. In this paper, among various gradual scene changes, we focus on dissolves. The algorithm uses a subset of the entire video, namely the sequence of DC images, for improvement of detection velocity

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2007.06a
• /
• pp.703-706
• /
• 2007

The real-time monitoring about the activity of the human provides useful information about the activity quantity and an ability. This study developed a system for human physical activity assessment in ambulatory monitoring using portable sensing device combining a tri-axial accelerometer and wireless sensor node. This real-time system is able to identify several postures, posture transitions and movements with classification algorithm. In addition, this system also features fall detection capability. The results of the assessment for evaluating the performance of the system show high identification accuracy.