Wearable Sensor Data Classification for Human Activity Recognition Based on an Iterative Learning Framework

Decoding of unimanual and bimanual reach-and-grasp actions from EMG and IMU signals in persons with cervical spinal cord injury

Objective. Chronic motor impairments of arms and hands as the consequence of a cervical spinal cord injury (SCI) have a tremendous impact on activities of daily life. A considerable number of people however retain minimal voluntary motor control in the paralyzed parts of the upper limbs that are measurable by electromyography (EMG) and inertial measurement units (IMUs). An integration into human-machine interfaces (HMIs) holds promise for reliable grasp intent detection and intuitive assistive device control.Approach. We used a multimodal HMI incorporating EMG and IMU data to decode reach-and-grasp movements of groups of persons with cervical SCI (n = 4) and without (control, n = 13). A post-hoc evaluation of control group data aimed to identify optimal parameters for online, co-adaptive closed-loop HMI sessions with persons with cervical SCI. We compared the performance of real-time, Random Forest-based movement versus rest (2 classes) and grasp type predictors (3 classes) with respect to their co-adaptation and evaluated the underlying feature importance maps.Main results. Our multimodal approach enabled grasp decoding significantly better than EMG or IMU data alone (p<0.05). We found the 0.25 s directly prior to the first touch of an object to hold the most discriminative information. Our HMIs correctly predicted 79.3 ± STD 7.4 (102.7 ± STD 2.3 control group) out of 105 trials with grand average movement vs. rest prediction accuracies above 99.64% (100% sensitivity) and grasp prediction accuracies of 75.39 ± STD 13.77% (97.66 ± STD 5.48% control group). Co-adaption led to higher prediction accuracies with time, and we could identify adaptions in feature importances unique to each participant with cervical SCI.Significance. Our findings foster the development of multimodal and adaptive HMIs to allow persons with cervical SCI the intuitive control of assistive devices to improve personal independence.

Research on the Application of Multi-Source Information Fusion in Multiple Gait Pattern Transition Recognition

Multi-source information fusion technology is a kind of information processing technology which comprehensively processes and utilizes multi-source uncertain information. It is an effective scheme to solve complex pattern recognition and improve classification performance. This study aims to improve the accuracy and robustness of exoskeleton gait pattern transition recognition in complex environments. Based on the theory of multi-source information fusion, this paper explored a multi-source information fusion model for exoskeleton gait pattern transition recognition in terms of two aspects of multi-source information fusion strategy and multi-classifier fusion. For eight common gait pattern transitions (between level and stair walking and between level and ramp walking), we proposed a hybrid fusion strategy of multi-source information at the feature level and decision level. We first selected an optimal feature subset through correlation feature extraction and feature selection algorithm, followed by the feature fusion through the classifier. We then studied the construction of a multi-classifier fusion model with a focus on the selection of base classifier and multi-classifier fusion algorithm. By analyzing the classification performance and robustness of the multi-classifier fusion model integrating multiple classifier combinations with a number of multi-classifier fusion algorithms, we finally constructed a multi-classifier fusion model based on D-S evidence theory and the combination of three SVM classifiers with different kernel functions (linear, RBF, polynomial). Such multi-source information fusion model improved the anti-interference and fault tolerance of the model through the hybrid fusion strategy of feature level and decision level and had higher accuracy and robustness in the gait pattern transition recognition, whose average recognition accuracy for eight gait pattern transitions reached 99.70%, which increased by 0.15% compared with the highest average recognition accuracy of the single classifier. Moreover, the average recognition accuracy in the absence of different feature data reached 97.47% with good robustness.

Development and validation of smartwatch-based activity recognition models for rigging crew workers on cable logging operations

Analysis of high-resolution inertial sensor and global navigation satellite system (GNSS) data collected by mobile and wearable devices is a relatively new methodology in forestry and safety research that provides opportunities for modeling work activities in greater detail than traditional time study analysis. The objective of this study was to evaluate whether smartwatch-based activity recognition models could quantify the activities of rigging crew workers setting and disconnecting log chokers on cable logging operations. Four productive cycle elements (travel to log, set choker, travel away, clear) were timed for choker setters and four productive cycle elements (travel to log, unhook, travel away, clear) were timed for chasers working at five logging sites in North Idaho. Each worker wore a smartwatch that recorded accelerometer data at 25 Hz. Random forest machine learning was used to develop predictive models that classified the different cycle elements based on features extracted from the smartwatch acceleration data using 15 sliding window sizes (1 to 15 s) and five window overlap levels (0%, 25%, 50%, 75%, and 90%). Models were compared using multiclass area under the Receiver Operating Characteristic (ROC) curve, or AUC. The best choker setter model was created using a 3-s window with 90% overlap and had sensitivity values ranging from 76.95% to 83.59% and precision values ranging from 41.42% to 97.08%. The best chaser model was created using a 1-s window with 90% overlap and had sensitivity values ranging from 71.95% to 82.75% and precision values ranging from 14.74% to 99.16%. These results have demonstrated the feasibility of quantifying forestry work activities using smartwatch-based activity recognition models, a basic step needed to develop real-time safety notifications associated with high-risk job functions and to advance subsequent, comparative analysis of health and safety metrics across stand, site, and work conditions.

A Study of Accelerometer and Gyroscope Measurements in Physical Life-Log Activities Detection Systems

Nowadays, wearable technology can enhance physical human life-log routines by shifting goals from merely counting steps to tackling significant healthcare challenges. Such wearable technology modules have presented opportunities to acquire important information about human activities in real-life environments. The purpose of this paper is to report on recent developments and to project future advances regarding wearable sensor systems for the sustainable monitoring and recording of human life-logs. On the basis of this survey, we propose a model that is designed to retrieve better information during physical activities in indoor and outdoor environments in order to improve the quality of life and to reduce risks. This model uses a fusion of both statistical and non-statistical features for the recognition of different activity patterns using wearable inertial sensors, i.e., triaxial accelerometers, gyroscopes and magnetometers. These features include signal magnitude, positive/negative peaks and position direction to explore signal orientation changes, position differentiation, temporal variation and optimal changes among coordinates. These features are processed by a genetic algorithm for the selection and classification of inertial signals to learn and recognize abnormal human movement. Our model was experimentally evaluated on four benchmark datasets: Intelligent Media Wearable Smart Home Activities (IM-WSHA), a self-annotated physical activities dataset, Wireless Sensor Data Mining (WISDM) with different sporting patterns from an IM-SB dataset and an SMotion dataset with different physical activities. Experimental results show that the proposed feature extraction strategy outperformed others, achieving an improved recognition accuracy of 81.92%, 95.37%, 90.17%, 94.58%, respectively, when IM-WSHA, WISDM, IM-SB and SMotion datasets were applied.

Supervised Machine Learning Applied to Wearable Sensor Data Can Accurately Classify Functional Fitness Exercises Within a Continuous Workout

Observing, classifying and assessing human movements is important in many applied fields, including human-computer interface, clinical assessment, activity monitoring and sports performance. The redundancy of options in planning and implementing motor programmes, the inter- and intra-individual variability in movement execution, and the time-continuous, high-dimensional nature of motion data make segmenting sequential movements into a smaller set of discrete classes of actions non-trivial. We aimed to develop and validate a method for the automatic classification of four popular functional fitness drills, which are commonly performed in current circuit training routines. Five inertial measurement units were located on the upper and lower limb, and on the trunk of fourteen participants. Positions were chosen by keeping into account the dynamics of the movement and the positions where commercially-available smart technologies are typically secured. Accelerations and angular velocities were acquired continuously from the units and used to train and test different supervised learning models, including k-Nearest Neighbors (kNN) and support-vector machine (SVM) algorithms. The use of different kernel functions, as well as different strategies to segment continuous inertial data were explored. Classification performance was assessed from both the training dataset (k-fold cross-validation), and a test dataset (leave-one-subject-out validation). Classification from different subsets of the measurement units was also evaluated (1-sensor and 2-sensor data). SVM with a cubic kernel and fed with data from 600 ms windows with a 10% overlap gave the best classification performances, yielding to an overall accuracy of 97.8%. This approach did not misclassify any functional fitness movement for another, but confused relatively frequently (2.8-18.9%) a fitness movement phase with the transition between subsequent repetitions of the same task or different drills. Among 1-sensor configurations, the upper arm achieved the best classification performance (96.4% accuracy), whereas combining the upper arm and the thigh sensors obtained the highest level of accuracy (97.6%) from 2-sensors movement tracking. We found that supervised learning can successfully classify complex sequential movements such as those of functional fitness workouts. Our approach, which could exploit technologies currently available in the consumer market, demonstrated exciting potential for future on-field applications including unstructured training.

The new design of cows' behavior classifier based on acceleration data and proposed feature set

Monitor and classify behavioral activities in cows is a helpful support solution for livestock based on the analysis of data from sensors attached to the animal. Accelerometers are particularly suited for monitoring cow behaviors due to small size, lightweight and high accuracy. Nevertheless, the interpretation of the data collected by such sensors when characterizing the type of behaviors still brings major challenges to developers, related to activity complexity (i.e., certain behaviors contain similar gestures). This paper presents a new design of cows' behavior classifier based on acceleration data and proposed feature set. Analysis of cow acceleration data is used to extract features for classification using machine learning algorithms. We found that with 5 features (mean, standard deviation, root mean square, median, range) and 16-second window of data (1 sample/second), classification of seven cow behaviors (including feeding, lying, standing, lying down, standing up, normal walking, active walking) achieved the overall highest performance. We validated the results with acceleration data from a public source. Performance of our proposed classifier was evaluated and compared to existing ones in terms of the sensitivity, the accuracy, the positive predictive value, and the negative predictive value.

A User-Adaptive Algorithm for Activity Recognition Based on K-Means Clustering, Local Outlier Factor, and Multivariate Gaussian Distribution

Mobile activity recognition is significant to the development of human-centric pervasive applications including elderly care, personalized recommendations, etc. Nevertheless, the distribution of inertial sensor data can be influenced to a great extent by varying users. This means that the performance of an activity recognition classifier trained by one user’s dataset will degenerate when transferred to others. In this study, we focus on building a personalized classifier to detect four categories of human activities: light intensity activity, moderate intensity activity, vigorous intensity activity, and fall. In order to solve the problem caused by different distributions of inertial sensor signals, a user-adaptive algorithm based on K-Means clustering, local outlier factor (LOF), and multivariate Gaussian distribution (MGD) is proposed. To automatically cluster and annotate a specific user’s activity data, an improved K-Means algorithm with a novel initialization method is designed. By quantifying the samples’ informative degree in a labeled individual dataset, the most profitable samples can be selected for activity recognition model adaption. Through experiments, we conclude that our proposed models can adapt to new users with good recognition performance.