Posture Classification Research Articles

Ultra-Wideband Radar-Based Indoor Activity Monitoring for Elderly Care.

In this paper, we propose an unobtrusive method and architecture for monitoring a person’s presence and collecting his/her health-related parameters simultaneously in a home environment. The system is based on using a single ultra-wideband (UWB) impulse-radar as a sensing device. Using UWB radars, we aim to recognize a person and some preselected movements without camera-type monitoring. Via the experimental work, we have also demonstrated that, by using a UWB signal, it is possible to detect small chest movements remotely to recognize coughing, for example. In addition, based on statistical data analysis, a person’s posture in a room can be recognized in a steady situation. In addition, we implemented a machine learning technique (k-nearest neighbour) to automatically classify a static posture using UWB radar data. Skewness, kurtosis and received power are used in posture classification during the postprocessing. The classification accuracy achieved is more than 99%. In this paper, we also present reliability and fault tolerance analyses for three kinds of UWB radar network architectures to point out the weakest item in the installation. This information is highly important in the system’s implementation.

Automated postural ergonomic risk assessment using vision-based posture classification

Construction workers are at high risk of work-related musculoskeletal disorders (WMSDs) due to physically demanding manual-handling tasks in awkward postures. Although existing observational methods to identify ergonomic risks are inexpensive and easy to use, they are seldom used in construction sites because they are time-consuming, subject to observer bias, and require well-trained analysts. To address these drawbacks, this paper proposes a vision-based method to automatically classify workers' postures for ergonomic assessment. Specifically, it proposes a vision-based method that eliminates the need to collect extensive training-image datasets by employing classification algorithms to learn diverse postures from virtual images, and then identifies those postures in real-world images. The experimental tests showed about 89% classification accuracy in automatically classifying diverse postures on images, confirming the usefulness of virtual training images for posture classification. The proposed method has potential for automated ergonomic risk analysis, and could help to prevent WMSDs during diverse occupational tasks.

A CNN-based posture change detection for lactating sow in untrimmed depth videos

The dangerous body movements of lactating sows are mainly reflected in the frequency and duration of posture changes. To avoid the disturbances from the light environment and illumination variations caused by heat lamp and day-night cycle, depth videos were acquired for posture change detection. In this paper, a CNN-based method was developed to automatically detect sow posture change in untrimmed depth videos. Firstly, clip-level Detect and Classify Posture R-CNN (DCP R-CNN) was proposed for sow detection and posture classification. The DCP R-CNN decoupled the detection and classification task into two branches: detection branch and classification branch so that the detection branch could focus on the task of sow localization. Secondly, a method was proposed to temporally localize the possible posture changing segments from coarse to fine. Finally, a two-stream posture change classification network, which used two CNNs to extract the complementary features from DMM and the first and the last frames, was introduced to classify the possible posture changing segments into change or non-change. At the scene of commercial piggery, the videos were acquired at 5 frames per second (fps) by installing a Microsoft Kinect v2.0 sensor above the pen. The test set consisted of two 24-hour untrimmed depth videos and six 40-minute untrimmed depth videos. The test results showed that our method could detect the posture change correctly. The success rate of DCP R-CNN was 96.38% when the IoU threshold was 0.7, and the average precision and recall rate of 4 types of posture were 97.72% and 95.23%, respectively. The possible posture change was detected with a recall rate of 92.46% and the posture change was classified with an accuracy of 97.71%. The precision of posture change detection was 88.75%, the recall rate was 90.87%. The detection speed of our method was 5.71 fps, which was faster than the video frame rate (5 fps). Our study demonstrated the feasibility of 24-hour detection method for sow posture change.

Monitoring of Sitting Postures With Sensor Networks in Controlled and Free-living Environments: Systematic Review.

A majority of employees in the industrial world spend most of their working time in a seated position. Monitoring sitting postures can provide insights into the underlying causes of occupational discomforts such as low back pain. This study focuses on the technologies and algorithms used to classify sitting postures on a chair with respect to spine and limb movements, using sensors and wearables such as inertial measurement units, pressure or piezoresistive sensors, accelerometers or gyroscopes, combined with machine learning approaches. A total of three electronic literature databases were surveyed to identify studies classifying sitting postures in adults. Quality appraisal was performed to extract critical details and assess biases in the shortlisted papers. A total of 14 papers were shortlisted from 952 papers obtained after a systematic search. The majority of the studies used pressure sensors to measure sitting postures, whereas neural networks were the most frequently used approaches for classification tasks in this context. Only 2 studies were performed in a free-living environment. Most studies presented ethical and methodological shortcomings. Moreover, the findings indicate that the strategic placement of sensors can lead to better performance and lower costs. The included studies differed in various aspects of design and analysis. The majority of studies were rated as medium quality according to our assessment. Our study suggests that future work for posture classification can benefit from using inertial measurement unit sensors, since they make it possible to differentiate among spine movements and similar postures, considering transitional movements between postures, and using three-dimensional cameras to annotate the data for ground truth. Finally, comparing such studies is challenging, as there are no standard definitions of sitting postures that could be used for classification. In addition, this study identifies five basic sitting postures along with different combinations of limb and spine movements to help guide future research efforts.

Detection of posture and mobility in individuals at risk of developing pressure ulcers

Pressure mapping technologies provide the opportunity to estimate trends in posture and mobility over extended periods in individuals at risk of developing pressure ulcers. The aim of the study was to combine pressure monitoring with an automated algorithm to detect posture and mobility in a vulnerable population of Spinal Cord Injured (SCI) patients.Pressure data from able-bodied cohort studies involving prescribed lying and sitting postures were used to train the algorithm. This was tested with data from two SCI patients. Variations in the trends of the centre of pressure (COP) and contact area were assessed for detection of small- and large-scale postural movements. Intelligent data processing involving a deep learning algorithm, namely a convolutional neural network (CNN), was utilised for posture classification.COP signals revealed perturbations indicative of postural movements, which were automatically detected using individual- and movement-specific thresholds. CNN provided classification of static postures, with an accuracy ranging between 70-84% in the training cohort of able-bodied subjects. A clinical evaluation highlighted the potential of the novel algorithm to detect postural movements and classify postures in SCI patients.Combination of continuous pressure monitoring and intelligent algorithms offers the potential to objectively detect posture and mobility in vulnerable patients and inform clinical-decision making to provide personalized care.

Textile-Based Pressure Sensing Matrix for In-Bed Monitoring of Subject Sleeping Posture and Breathing Activity

According to current trends in healthcare sensing technologies, we describe a textile-based pressure sensing matrix that can be integrated in the mattress of a smart bed to characterize sleeping posture/movement of a subject and to extract breathing activity. The pressure mapping layer is developed as a matrix of 195 piezoresistive sensors, it is entirely made of textile materials, and it is the basic component of a smart bed that can perform sleep analysis, can extract physiological parameters, and can detect environmental data related to subject’s health. In this paper, we show the principle of the pressure mapping layer and the architecture of the dedicated electronic system that we developed for signal acquisition. In addition, we describe the algorithms for posture/movement classification (dedicated artificial neural network) and for extraction of the breathing rate (frequency domain analysis). We also perform validation of the system to quantify the accuracy/precision of the posture classification and the statistical analysis to compare our breathing rate estimation with the gold standard.

A computer vision-based approach for behavior recognition of gestating sows fed different fiber levels during high ambient temperature.

The objectives of this study were to evaluate convolutional neural network models and computer vision techniques for the classification of swine posture with high accuracy and to use the derived result in the investigation of the effect of dietary fiber level on the behavioral characteristics of the pregnant sow under low and high ambient temperatures during the last stage of gestation. A total of 27 crossbred sows (Yorkshire × Landrace; average body weight, 192.2 ± 4.8 kg) were assigned to three treatments in a randomized complete block design during the last stage of gestation (days 90 to 114). The sows in group 1 were fed a 3% fiber diet under neutral ambient temperature; the sows in group 2 were fed a diet with 3% fiber under high ambient temperature (HT); the sows in group 3 were fed a 6% fiber diet under HT. Eight popular deep learning-based feature extraction frameworks (DenseNet121, DenseNet201, InceptionResNetV2, InceptionV3, MobileNet, VGG16, VGG19, and Xception) used for automatic swine posture classification were selected and compared using the swine posture image dataset that was constructed under real swine farm conditions. The neural network models showed excellent performance on previously unseen data (ability to generalize). The DenseNet121 feature extractor achieved the best performance with 99.83% accuracy, and both DenseNet201 and MobileNet showed an accuracy of 99.77% for the classification of the image dataset. The behavior of sows classified by the DenseNet121 feature extractor showed that the HT in our study reduced (p < 0.05) the standing behavior of sows and also has a tendency to increase (p = 0.082) lying behavior. High dietary fiber treatment tended to increase (p = 0.064) lying and decrease (p < 0.05) the standing behavior of sows, but there was no change in sitting under HT conditions.

Fall Detection System-Based Posture-Recognition for Indoor Environments.

The majority of the senior population lives alone at home. Falls can cause serious injuries, such as fractures or head injuries. These injuries can be an obstacle for a person to move around and normally practice his daily activities. Some of these injuries can lead to a risk of death if not handled urgently. In this paper, we propose a fall detection system for elderly people based on their postures. The postures are recognized from the human silhouette which is an advantage to preserve the privacy of the elderly. The effectiveness of our approach is demonstrated on two well-known datasets for human posture classification and three public datasets for fall detection, using a Support-Vector Machine (SVM) classifier. The experimental results show that our method can not only achieves a high fall detection rate but also a low false detection.

High-throughput behavioral screen in C. elegans reveals Parkinson\u2019s disease drug candidates

We recently linked branched-chain amino acid transferase 1 (BCAT1) dysfunction with the movement disorder Parkinson’s disease (PD), and found that RNAi-mediated knockdown of neuronal bcat-1 in C. elegans causes abnormal spasm-like ‘curling’ behavior with age. Here we report the development of a machine learning-based workflow and its application to the discovery of potentially new therapeutics for PD. In addition to simplifying quantification and maintaining a low data overhead, our simple segment-train-quantify platform enables fully automated scoring of image stills upon training of a convolutional neural network. We have trained a highly reliable neural network for the detection and classification of worm postures in order to carry out high-throughput curling analysis without the need for user intervention or post-inspection. In a proof-of-concept screen of 50 FDA-approved drugs, enasidenib, ethosuximide, metformin, and nitisinone were identified as candidates for potential late-in-life intervention in PD. These findings point to the utility of our high-throughput platform for automated scoring of worm postures and in particular, the discovery of potential candidate treatments for PD.

OWAS inter-rater reliability

The Ovako Working posture Assessment System (OWAS) is a commonly used observational assessment method for determining the risk of work-related musculoskeletal disorders. OWAS claims to be suitable in the application for untrained persons but there is not enough evidence for this assumption. In this paper, inter-rater (inter-observer) reliability (agreement) is examined down to the level of individual postures and categories. For this purpose, the postures of 20 volunteers have been observed by 3 varying human raters in a laboratory setting and the inter-rater agreement against reference values was determined. A high agreement of over 98%(κ=0.98) was found for the postures of the arms but lower agreements were found for posture classification of the legs (66−97%,κ=0.85) and the upper body (80−96%,κ=0.85). No significant difference was found between raters with and without intense prior training in physical therapy. Consequently, the results confirm the general reliability of the OWAS method especially for raters with non-specialized background but suggests weaknesses in the reliable detection of a few particular postures.

Vision Based Detection of Driver Cell Phone Usage and Food Consumption

Distracted driving is a problem which yearly causes a large amount of road traffic crashes with high rates of fatalities and injured persons. Recently, car manufacturers started to integrate driver monitoring systems to detect visual distraction. This paper proposes a method to extend such systems by driver posture classification to detect driver cell phone usage and food consumption. Such an extension can be beneficial since systems that focus on the detection of visual distraction mainly rely on head pose and gaze information. Thus, distraction caused by cell phone usage or food consumption can not be detected by these systems when the driver is looking to the road ahead. To robustly detect those types of manual and cognitive distraction, different deep learning models were trained and evaluated based on a new image dataset which was captured by two infrared cameras to ensure that a large range of head angles can be covered by the system. Separate Convolutional Neural Networks (CNNs) were trained and evaluated for the dataset of the left and the right camera to optimize the classification accuracy. The trained CNNs revealed a competitive test accuracy of 92.88% and 90.36% for the left and the right camera, respectively. In inference mode, the models achieve a frame rate of 44Hz and 28Hz for the left and the right camera, respectively. The combination of the classification output of both networks revealed a test accuracy of 92.54%.

Classification of high knee flexion postures using EMG signals.

High knee flexion postures are often adopted in occupational settings and may lead to increased risk of knee osteoarthritis. Pattern recognition algorithms using wireless electromyographic (EMG) signals may be capable of detecting and quantifying occupational exposures throughout a working day. To develop a k-Nearest Neighbor (kNN) algorithm for the classification of eight high knee flexion activities frequently observed in childcare. EMG signals from eight lower limb muscles were recorded for 30 participants, signals were decomposed into time- and frequency-domain features, and used to develop a kNN classification algorithm. Features were reduced to a combination of ten time-domain features from 8 muscles using neighborhood component analysis, in order to most effectively identify the postures of interest. The final classifier was capable of accurately identifying 80.1%of high knee flexion postures based on novel data from participants included in the training dataset, yet only achieved 18.4%accuracy when predicting postures based on novel subject data. EMG based classification of high flexion postures may be possible within occupational settings when the model is first trained on sample data from a given individual. The developed algorithm may provide quantitative measures leading to a greater understanding of occupation specific postural requirements.

Where to Place Which Sensor to Measure Sedentary Behavior? A Method Development and Comparison Among Various Sensor Placements and Signal Types

Background: Sedentary behavior (SB) is associated with several chronic diseases and office workers especially are at increased risk. SB is defined by a sitting or reclined body posture with an energy expenditure of ≤1.5 metabolic equivalents. However, current objective methods to measure SB are not consistent with its definition. There is no consensus on which sensor placement and type should be used. Aim: To compare the accuracy of newly developed artificial intelligence models for 15 sensor placements in combination with four signal types (accelerometer only/plus gyroscope and/or magnetometer) to detect posture and physical in-/activity during desk-based activities. Method: Signal features for the model development were extracted from sensor raw data of 30 office workers performing 10 desk-based tasks, each lasting 5 min. Direct observation (posture) and indirect calorimetry (in-/activity) served as the reference criteria. The best classification model for each sensor was identified and compared among the sensor placements, both using Friedman and post hoc Wilcoxon tests (p ≤ .05). Results: Posture was most accurately measured with a lower body sensor, while in-/activity was most accurately measured with an upper body or waist sensor. The inclusion of additional signal types improved the posture classification for some placements, while the acceleration signal already contained the relevant signal information for the in-/activity classification. Overall, the thigh accelerometer most accurately classified desk-based SB. Conclusion: This study favors, in line with previous work, the measurement of SB with a thigh-worn accelerometer and adds the information that this sensor is also accurate in measuring physical in-/activity while sitting and standing.

Inter-tester Reliability of Lumbar Lordosis Posture Classification Using a Novel Screening Device

ObjectiveThe purpose of this study was to investigate the inter-tester reliability of lumbar lordosis posture using a novel screening device. MethodsA total of 33 healthy young male participants participated in the study. Two examiners measured the regional upper and lower lumbar lordosis angles of the participants in the standing position using a flexible ruler. The bent flexible ruler maintained a fixed shape and was transferred to a protractor for angle measurement. Two examiners classified each participant into one of 4 lumbar spine categories and measured the upper and lower regional lumbar lordosis angles. ResultsThe agreement level between the 2 examiners in assessing healthy participants was 87.9%. The calculated kappa coefficient was 0.79 (95% CI = 0.86-0.97), reflecting a substantial level of agreement. ConclusionOur results suggest that our novel screening device for assessing upper and lower lumbar angles showed good inter-tester reliability in posture classification. Our findings may be useful for health care professionals for managing sagittal lumbar posture in asymptomatic younger individuals; however, more testing is still needed.

Comprehensive Monitoring of Bad Head and Shoulder Postures by Wearable Magnetic Sensors and Deep Learning

In this work, a simultaneous monitoring method for bad posture, including the forward head posture (FHP), rounded shoulder (RS), and elevated shoulder (ES), is proposed. These postures and the resulting symptoms are becoming increasingly prevalent, and a comprehensive, simultaneous, and extensive analysis of such posture disorders is needed. The proposed method involves collecting posture data from a new combination of accelerometers and magnetometers paired with miniature magnets. The sensor locations are optimally chosen to reliably calculate neck and shoulder angles representing the craniovertebral angle (CVA) for FHP, forward shoulder angle (FSA) for RS, and symmetry angle (SA) for ES. Processing of the collected sensor data is achieved by deep neural network (DNN) and convolutional neural network (CNN) algorithms. Experimental results demonstrate successful bad posture classification with high accuracy (DNN: 88.1%, CNN: 88.7%) even with the simultaneous analysis of FHP, RS, and ES.

지각 불편도에 근거한 복합 목 동작 분류 체계

Objective: The objectives of this study are to investigate effects of complex neck motions including flexion/extension, rotation, and lateral bending on perceived discomfort, and to develop a new posture classification scheme for complex neck motions based on the perceived discomfort.BRBRBackground: As the use of computers and smart phones is increased, neck relevant disorders are increasing. For preventing or reducing the disorders, postural load for neck motions should be precisely quantified. There is a lack of study or assessment tool for complex neck motions.BRBRMethod: An experiment for measuring perceived discomfort according to complex neck motions was performed, in which neck motions of flexion/extension, rotation, and lateral bending were used as independent variables. Discomfort according to complex neck motions was used as dependent variable and measured by using the Borg CR10. Sixteen healthy male college students without histories of musculoskeletal disorders participated in the experiment.BRBRResults: ANOVA revealed that while the interactions between the independent variables were not significant, single factors of flexion/extension, rotation, and lateral bending affected significantly on the discomfort at α = 0.01. Based on the results, a new posture classification scheme for complex neck motions was developed and validated.BRBRConclusion: This study investigated effects of complex neck motions on perceived discomfort, and developed a new classification scheme of complex neck motions.BRBRApplication: The results of this study will help more precisely assess postural load according to complex neck motions as well as to develop a new posture classification scheme.

Pervasive Lying Posture Tracking.

Automated lying-posture tracking is important in preventing bed-related disorders, such as pressure injuries, sleep apnea, and lower-back pain. Prior research studied in-bed lying posture tracking using sensors of different modalities (e.g., accelerometer and pressure sensors). However, there remain significant gaps in research regarding how to design efficient in-bed lying posture tracking systems. These gaps can be articulated through several research questions, as follows. First, can we design a single-sensor, pervasive, and inexpensive system that can accurately detect lying postures? Second, what computational models are most effective in the accurate detection of lying postures? Finally, what physical configuration of the sensor system is most effective for lying posture tracking? To answer these important research questions, in this article we propose a comprehensive approach for designing a sensor system that uses a single accelerometer along with machine learning algorithms for in-bed lying posture classification. We design two categories of machine learning algorithms based on deep learning and traditional classification with handcrafted features to detect lying postures. We also investigate what wearing sites are the most effective in the accurate detection of lying postures. We extensively evaluate the performance of the proposed algorithms on nine different body locations and four human lying postures using two datasets. Our results show that a system with a single accelerometer can be used with either deep learning or traditional classifiers to accurately detect lying postures. The best models in our approach achieve an score that ranges from % to % with a coefficient of variation from to . The results also identify the thighs and chest as the most salient body sites for lying posture tracking. Our findings in this article suggest that, because accelerometers are ubiquitous and inexpensive sensors, they can be a viable source of information for pervasive monitoring of in-bed postures.

Developing and Evaluating a Mixed Sensor Smart Chair System for Real-Time Posture Classification: Combining Pressure and Distance Sensors.

A novel sensor-embedded smart chair system was developed to monitor and classify a worker's sitting postures in real time. The smart chair system was a mixed sensor system utilizing six pressure sensors and six infrared reflective distance sensors in combination. The pressure sensors were embedded in the seat cushion to gather seat cushion pressure distribution data. The distance sensors were placed in the seatback to measure seatback-trunk distances at different locations in the frontal plane. The use of the seatback distance sensors represented a unique design feature, which distinguished the mixed sensor system from the previous posture monitoring systems. Employing a k-Nearest Neighbor algorithm, the mixed sensor system classified an instantaneous posture as one of posture categories determined based on an analysis of the ergonomics literature on sitting postures and sitting-related musculoskeletal problems. The mixed sensor system was evaluated in posture classification performance in comparison with two benchmark systems that utilized only a single type of sensors. The purpose of the comparisons was to determine the utility of the design combining seat cushion pressure sensors and seatback distance sensors. The mixed sensor system yielded significantly superior classification performance than the two benchmark systems. The mixed sensor system is low-cost utilizing only a small number of sensors; yet, it accomplishes accurate classification of postures relevant to the ergonomic analyses of seated work tasks. The mixed sensor system could be utilized for various applications including the development of a real-time posture feedback system for preventing sitting-related musculoskeletal disorders.

Using machine learning for in-bed posture classification, sleep and daily pattern recognition on a motion-sensing mattress

Using machine learning for in-bed posture classification, sleep and daily pattern recognition on a motion-sensing mattress

Hand posture classification with convolutional neural networks on VGG-19 net Architecture

This study aims to classify the image depth data Hand Posture. Hand Posture is a form of hand and movement used to communicate. Hand Posture is difficult to classify because various human hand objects are complex articulation objects. The model used in this study is Convolutional Neural Networks using the VGG-19 Net architecture. Based on the results shows an increase in the percentage of classification accuracy in each subject is 0.9976, 1.0, 0.9984, 1.0, and 0.9992 respectively.