Accelerometer-Based Activity Recognition in Construction
01 Sep 2011-Journal of Computing in Civil Engineering (American Society of Civil Engineers)-Vol. 25, Iss: 5, pp 370-379
TL;DR: The study showed that the utilization of best features instead of all features did not affect the classification accuracy significantly but reduced the run time considerably, and multilayer perceptron, a neural network classifier, gave the best results.
Abstract: Recognizing the activities of workers helps to measure and control safety, productivity, and quality in construction sites. Automated activity recognition can enhance the efficiency of the measurement system. The present study investigates accelerometer-based activity classification for automating the work-sampling process. A methodology is developed for evaluating classifiers for recognizing activities based on the features generated from accelerometer data segments. An experimental study is carried out in instructed and uninstructed modes for classifying masonry activities by using accelerometers attached to the waist of the mason. Three types of classifiers were evaluated, and multilayer perceptron, a neural network classifier, gave the best results. A 50% overlap for data segments enhanced classifier performance. The study showed that the utilization of best features instead of all features did not affect the classification accuracy significantly but reduced the run time considerably. An accuracy of 8...
Citations
More filters
TL;DR: Smartphones are used in an unobtrusive way to capture body movements by collecting data using embedded accelerometer and gyroscope sensors and results indicate that neural networks outperform other classifiers by offering an accuracy ranging from 87% to 97% for user-dependent and 62% to 96% foruser-independent categories.
Abstract: Understanding the state, behavior, and surrounding context of construction workers is essential to effective project management and control. Exploiting the integrated sensors of ubiquitous mobile phones offers an unprecedented opportunity for an automated approach to workers' activity recognition. In addition, machine learning (ML) methodologies provide the complementary computational part of the process. In this paper, smartphones are used in an unobtrusive way to capture body movements by collecting data using embedded accelerometer and gyroscope sensors. Construction activities of various types have been simulated and collected data are used to train five different types of ML algorithms. Activity recognition accuracy analysis has been performed for all the different categories of activities and ML classifiers in user-dependent and -independent ways. Results indicate that neural networks outperform other classifiers by offering an accuracy ranging from 87% to 97% for user-dependent and 62% to 96% for user-independent categories.
185 citations
Cites background from "Accelerometer-Based Activity Recogn..."
...Joshua and Varghese [7] adopted a similar approach to facilitate the manual process of work sampling in construction projects. khavian), Process monitoring and control provides a solid basis for tracking and measurements required for activity analysis....
[...]
...Joshua and Varghese [7] adopted a similar approach to facilitate the manual process of work sampling in construction projects....
[...]
...Joshua and Varghese [7] were among the first researchers who explored the application of accelerometer in construction for work sampling....
[...]
TL;DR: Work-related musculoskeletal disorders (WMSDs) have long been a primary cause of non-fatal injuries in construction and have been classified into self-report, observation, direct measurement, and remote sensing assessment as mentioned in this paper.
Abstract: Work-related musculoskeletal disorders (WMSDs) have long been a primary cause of non-fatal injuries in construction. They involve sudden or continuous stresses on a worker’s musculoskeletal system (e.g., muscles, tendons, ligaments, bones) and may impair the ability of the worker to perform his or her job, or even cause permanent disability. Although assessing exposure to risk factors of WMSDs has proven to be feasible to alleviate the incidence rate of this injury, the field remains underdeveloped because of a lack of knowledge among construction professionals regarding the enabling techniques and their performance and limits. This paper reviews the available techniques for WMSD risk assessments, summarizes their benefits and limitations, and identifies areas in which further studies are still needed. Current techniques are categorized into self-report, observation, direct measurement, and remote sensing assessment. Particular interests are revealed in the wearable-sensor and vision-based techniq...
165 citations
TL;DR: An in-depth review of existing assessment frameworks used in practice for the evaluation of human body motion and a new system to detect and characterise unsafe postures of construction workers based on the measurement of motion data from wearable wireless IMUs integrated in a body area network are introduced.
Abstract: Human body motions have been analysed for decades with a view on enhancing occupational well-being and performance of workers. On-going progresses in miniaturised wearable sensors are set to revolutionise biomechanical analysis by providing accurate and real-time quantitative motion data. The construction industry has a poor record of occupational health, in particular with regard to work-related musculoskeletal disorders (WMSDs). In this article, we therefore focus on the study of human body motions that could cause WMSDs in construction-related activities. We first present an in-depth review of existing assessment frameworks used in practice for the evaluation of human body motion. Subsequently different methods for measuring working postures and motions are reviewed and compared, pointing out the technological developments, limitations and gaps; Inertial Measurement Units (IMUs) are particularly investigated. Finally, we introduce a new system to detect and characterise unsafe postures of construction workers based on the measurement of motion data from wearable wireless IMUs integrated in a body area network. The potential of this system is demonstrated through experiments conducts in a laboratory as well as in a college with actual construction trade trainees.
163 citations
Cites background or result from "Accelerometer-Based Activity Recogn..."
...Although focused361 on productivity assesssment, Joshua and Varghese (2011a) present a system362 that classifies masonry activities (fetch and spread mortar, lay bricks, and363 filling joints) by processing acceleration data from two accelerometers placed364 on the waist of bricklayers....
[...]
...…EMG systems may be considered as intrusive341 (as noted above) and cannot be used outside the laboratory set up or in a342 real work site.343 Joshua and Varghese (2011b) propose to use video cameras to record the344 movement of construction workers on site and conduct an initial study of345…...
[...]
...In that sense, our system resembles398 that of Parkka et al. (2006), but their focus is on sport biomechanics while399 we focus on worker biomechanics for WMSDs risk assessment.400 This work also differs from that of Joshua and Varghese (2011a, 2014) who401 focus on productivity....
[...]
TL;DR: The goal of this research is to investigate the prospect of using built-in smartphone sensors as ubiquitous multi-modal data collection and transmission nodes in order to detect detailed construction equipment activities which can ultimately contribute to the process of simulation input modeling.
Abstract: Although activity recognition is an emerging general area of research in computer science, its potential in construction engineering and management (CEM) domain has not yet been fully investigated. Due to the complex and dynamic nature of many construction and infrastructure projects, the ability to detect and classify key activities performed in the field by various equipment and human crew can improve the quality and reliability of project decision-making and control. In particular to simulation modeling, process-level knowledge obtained as a result of activity recognition can help verify and update the input parameters of simulation models. Such input parameters include but are not limited to activity durations and precedence, resource flows, and site layout. The goal of this research is to investigate the prospect of using built-in smartphone sensors as ubiquitous multi-modal data collection and transmission nodes in order to detect detailed construction equipment activities which can ultimately contribute to the process of simulation input modeling. A case study of front-end loader activity recognition is presented to describe the methodology for action recognition and evaluate the performance of the developed system. In the designed methodology, certain key features are extracted from the collected data using accelerometer and gyroscope sensors, and a subset of the extracted features is used to train supervised machine learning classifiers. In doing so, several important technical details such as selection of discriminating features to extract, sensitivity analysis of data segmentation window size, and choice of the classifier to be trained are investigated. It is shown that the choice of the level of detail (LoD) in describing equipment actions (classes) is an important factor with major impact on the classification performance. Results also indicate that although decreasing the number of classes generally improves the classification output, considering other factors such as actions to be combined as a single activity, methodologies to extract knowledge from classified activities, computational efficiency, and end use of the classification process may as well influence one's decision in selecting an optimal LoD in describing equipment activities (classes).
157 citations
Cites methods from "Accelerometer-Based Activity Recogn..."
...Construction labor activity classification was also investigated to automate the work-sampling process [48]....
[...]
TL;DR: A low-cost ubiquitous approach is presented and validated which deploys built-in smartphone sensors to unobtrusively monitor workers' bodily postures and autonomously identify potential work-related ergonomic risks and indicates that measurements of trunk and shoulder flexions of a worker by smartphone sensory data are very close to corresponding measurements by observation.
Abstract: Construction jobs are more labor-intensive compared to other industries. As such, construction workers are often required to exceed their natural physical capability to cope with the increasing complexity and challenges in this industry. Over long periods of time, this sustained physical labor causes bodily injuries to the workers which in turn, conveys huge losses to the industry in terms of money, time, and productivity. Various safety and health organizations have established rules and regulations that limit the amount and intensity of workers' physical movements to mitigate work-related bodily injuries. A precursor to enforcing and implementing such regulations and improving the ergonomics conditions on the jobsite is to identify physical risks associated with a particular task. Manually assessing a field activity to identify the ergonomic risks is not trivial and often requires extra effort which may render it to be challenging if not impossible. In this paper, a low-cost ubiquitous approach is presented and validated which deploys built-in smartphone sensors to unobtrusively monitor workers' bodily postures and autonomously identify potential work-related ergonomic risks. Results indicates that measurements of trunk and shoulder flexions of a worker by smartphone sensory data are very close to corresponding measurements by observation. The proposed method is applicable for workers in various occupations who are exposed to WMSDs due to awkward postures. Examples include, but are not limited to industry laborers, carpenters, welders, farmers, health assistants, teachers, and office workers.
142 citations
Cites background from "Accelerometer-Based Activity Recogn..."
...Previous research has shown that these two positions produce the most distinctive features for most manual jobs performed by field workers (Joshua and Varghese, 2011; Yang and Hsu, 2010)....
[...]
References
More filters
Proceedings Article•
20 Aug 1995TL;DR: The results indicate that for real-word datasets similar to the authors', the best method to use for model selection is ten fold stratified cross validation even if computation power allows using more folds.
Abstract: We review accuracy estimation methods and compare the two most common methods crossvalidation and bootstrap. Recent experimental results on artificial data and theoretical re cults in restricted settings have shown that for selecting a good classifier from a set of classifiers (model selection), ten-fold cross-validation may be better than the more expensive leaveone-out cross-validation. We report on a largescale experiment--over half a million runs of C4.5 and a Naive-Bayes algorithm--to estimate the effects of different parameters on these algrithms on real-world datasets. For crossvalidation we vary the number of folds and whether the folds are stratified or not, for bootstrap, we vary the number of bootstrap samples. Our results indicate that for real-word datasets similar to ours, The best method to use for model selection is ten fold stratified cross validation even if computation power allows using more folds.
11,185 citations
21 Apr 2004
TL;DR: This is the first work to investigate performance of recognition algorithms with multiple, wire-free accelerometers on 20 activities using datasets annotated by the subjects themselves, and suggests that multiple accelerometers aid in recognition.
Abstract: In this work, algorithms are developed and evaluated to de- tect physical activities from data acquired using five small biaxial ac- celerometers worn simultaneously on different parts of the body. Ac- celeration data was collected from 20 subjects without researcher su- pervision or observation. Subjects were asked to perform a sequence of everyday tasks but not told specifically where or how to do them. Mean, energy, frequency-domain entropy, and correlation of acceleration data was calculated and several classifiers using these features were tested. De- cision tree classifiers showed the best performance recognizing everyday activities with an overall accuracy rate of 84%. The results show that although some activities are recognized well with subject-independent training data, others appear to require subject-specific training data. The results suggest that multiple accelerometers aid in recognition because conjunctions in acceleration feature values can effectively discriminate many activities. With just two biaxial accelerometers - thigh and wrist - the recognition performance dropped only slightly. This is the first work to investigate performance of recognition algorithms with multiple, wire-free accelerometers on 20 activities using datasets annotated by the subjects themselves.
3,223 citations
"Accelerometer-Based Activity Recogn..." refers background or methods or result in this paper
...Extending activity recognition to real-world situations was carried out by Bao and Intille (2004). Classification accuracies with various machine learning algorithms were compared for classifying 20 everyday activities like working on a computer, climbing stairs, vacuuming, brushing teeth, and folding laundry by using data collected from five biaxial accelerometers placed on different parts of the subject’s body....
[...]
...However, it was observed that 50% overlap in segments improved the performance, which is similar to results reported in past studies (Bao and Intille 2004)....
[...]
...In one such study, Bouten et al. (1997) recommended the position of waist or back as the appropriate place for attaching the sensor to minimize the influence of gravitational force on the accelerometer output....
[...]
Proceedings Article•
09 Jul 2005TL;DR: This paper reports on the efforts to recognize user activity from accelerometer data and performance of base-level and meta-level classifiers, and Plurality Voting is found to perform consistently well across different settings.
Abstract: Activity recognition fits within the bigger framework of context awareness. In this paper, we report on our efforts to recognize user activity from accelerometer data. Activity recognition is formulated as a classification problem. Performance of base-level classifiers and meta-level classifiers is compared. Plurality Voting is found to perform consistently well across different settings.
1,561 citations
01 Jan 2006
TL;DR: Results demonstrate the feasibility of implementing an accelerometry-based, real-time movement classifier using embedded intelligence.
Abstract: The real-time monitoring of human movement can provide valuable information regarding an individual's degree of functional ability and general level of activity. This paper presents the implementation of a real-time classification system for the types of human movement associated with the data acquired from a single, waist-mounted triaxial accelerometer unit. The major advance proposed by the system is to perform the vast majority of signal processing onboard the wearable unit using embedded intelligence. In this way, the system distinguishes between periods of activity and rest, recognizes the postural orientation of the wearer, detects events such as walking and falls, and provides an estimation of metabolic energy expenditure. A laboratory-based trial involving six subjects was undertaken, with results indicating an overall accuracy of 90.8% across a series of 12 tasks (283 tests) involving a variety of movements related to normal daily activities. Distinction between activity and rest was performed without error; recognition of postural orientation was carried out with 94.1% accuracy, classification of walking was achieved with less certainty (83.3% accuracy), and detection of possible falls was made with 95.6% accuracy. Results demonstrate the feasibility of implementing an accelerometry-based, real-time movement classifier using embedded intelligence
1,334 citations
Additional excerpts
...6% (Karantonis et al. 2006)....
[...]
TL;DR: Preliminary evaluation of the system in 13 male subjects during standardized activities in the laboratory demonstrated a significant relationship between accelerometer output and energy expenditure due to physical activity, the standard reference for physical activity.
Abstract: The present study describes the development of a triaxial accelerometer (TA) and a portable data processing unit for the assessment of daily physical activity. The TA is composed of three orthogonally mounted uniaxial piezoresistive accelerometers and can be used to register accelerations covering the amplitude and frequency ranges of human body acceleration. Interinstrument and test-retest experiments showed that the offset and the sensitivity of the TA were equal for each measurement direction and remained constant on two measurement days. Transverse sensitivity was significantly different for each measurement direction, but did not influence accelerometer output (<3% of the sensitivity along the main axis). The data unit enables the on-line processing of accelerometer output to a reliable estimator of physical activity over eight-day periods. Preliminary evaluation of the system in 13 male subjects during standardized activities in the laboratory demonstrated a significant relationship between accelerometer output and energy expenditure due to physical activity, the standard reference for physical activity (r=0.89). Shortcomings of the system are its low sensitivity to sedentary activities and the inability to register static exercise. The validity of the system for the assessment of normal daily physical activity and specific activities outside the laboratory should be studied in free-living subjects.
951 citations
"Accelerometer-Based Activity Recogn..." refers background or methods in this paper
...The generic concept of measurement of body movements with an accelerometer is discussed by Godfrey et al. (2008) with reference to past studies and modern techniques....
[...]
...Past studies have shown that the acceleration at body limbs and extremities can exhibit a 12 g range, whereas torso and hip experience 6 g range in acceleration (Bouten et al. 1997)....
[...]
...Past studies (Bouten et al. 1997) have shown that the amplitude spectra of human body accelerations at the waist level are in the range of 6 g to þ6 g, and hence, the range of the sensor was selected to be 6 g....
[...]