Search

Nowadays, for controlling crime, surveillance cameras are typically installed in all public places to ensure urban safety and security. However, automating Human Activity Recognition (HAR) using computer vision techniques faces several challenges such as lowlighting, complex spatiotemporal features, clutter backgrounds, and inefficient utilization of surveillance system resources. Existing attempts in HAR designed straightforward networks by analyzing either spatial or motion patterns resulting in limited performance while the dual streams methods are entirely based on Convolutional Neural Networks (CNN) that are inadequate to learning the long-range temporal information for HAR. To overcome the above-mentioned challenges, this paper proposes an optimized dual stream framework for HAR which mainly consists of three steps. First, a shots segmentation module is introduced in the proposed framework to efficiently utilize the surveillance system resources by enhancing the lowlight video stream and then it detects salient video frames that consist of human. This module is trained on our own challenging Lowlight Human Surveillance Dataset (LHSD) which consists of both normal and different levels of lowlighting data to recognize humans in complex uncertain environments. Next, to learn HAR from both contextual and motion information, a dual stream approach is used in the feature extraction. In the first stream, it freezes the learned weights of the backbone Vision Transformer (ViT) B-16 model to select the discriminative contextual information. In the second stream, ViT features are then fused with the intermediate encoder layers of FlowNet2 model for optical flow to extract a robust motion feature vector. Finally, a two stream Parallel Bidirectional Long Short-Term Memory (PBiLSTM) is proposed for sequence learning to capture the global semantics of activities, followed by Dual Stream Multi-Head Attention (DSMHA) with a late fusion strategy to optimize the huge features vector for accurate HAR. To assess the strength of the proposed framework, extensive empirical results are conducted on real-world surveillance scenarios and various benchmark HAR datasets that achieve 78.6285%, 96.0151%, and 98.875% accuracies on HMDB51, UCF101, and YouTube Action, respectively. Our results show that the proposed strategy outperforms State-of-the-Art (SOTA) methods. The proposed framework gives superior performance in HAR, providing accurate and reliable recognition of human activities in surveillance systems. [ABSTRACT FROM AUTHOR]

The construction industry still suffers from a low degree of transparency and bears health and safety risks for workers. A well-known disease is the Hand-Arm Vibration Syndrome, which results from the use of high-vibration tools. The recognition of worker activities to calculate a worker's exposure to vibration while operating tools in different modes has received little attention so far. This paper presents a machine-learning-based approach to recognize worker activities involving high-vibration hand-held power tools using acceleration data recorded with a smart-watch. The data is collected in a laboratory utilizing a combi-hammer in various tool settings. A Decision Tree and a Convolutional Pooling Long Short Term Memory (ConvPoolLSTM) model are developed to recognize worker activities on different levels. The ConvPoolLSTM architecture achieves accuracies between 89.1% and 96.1%. This research contributes towards the development of a more accurate, automated and low-cost monitoring system for vibration exposure time calculation and work process monitoring. • Machine learning models can recognize worker activities including power tools. • Collected data includes 202 min of smart-watch sensor data in various settings. • Models show high performance to recognize activities on three levels of details. • The ConvPoolLSTM model achieves up to 96.1% accuracy for tool runtime recognition. • The models can enable automated vibration time calculation on tool usage mode level. [ABSTRACT FROM AUTHOR]

Conducting an analysis of human behavior is an intriguing topic for many researchers. Within this field, machine learning can be applied to classify activities and emotions by analyzing physiological signals. However, the limited size of available databases poses challenges for the generalization of classifiers. This paper proposes a method to enhance the generalization of neural network-based classifiers by intelligently initializing weights for emotion and activity recognition. The signals under consideration are electrocardiogram, thoracic electrical bioimpedance, and electrodermal activity. The database used comprises recordings from 40 subjects performing various tasks that induce emotions and activities. The performance of the proposed method is compared with several standard machine learning and deep learning classifiers typically employed in emotion and activity recognition. This study involves two primary assessments. First is the activity recognition task, encompassing classes such as neutral, emotional, mental, and physical activity, where results close to 20% accuracy are achieved using the three physiological signals. Second, the emotion recognition assessment aims to differentiate between emotions like neutral, sadness, and disgust. An error probability close to 15% is obtained using thoracic electrical bioimpedance and electrodermal activity. The proposed method yields the best results among the approaches evaluated. • Database size limits the classification of the activity and emotions from biosignals. • Smart initialization of the weights in neural networks improves the generalization. • Our artificial network pre-training reduces local minima problems and overfitting. • The method is compared with several machine and deep learning classifiers. [ABSTRACT FROM AUTHOR]

Past research on activity recognition of heavy earth-moving machinery is limited to the statistical features of vibration signals with limited accuracy. Therefore, this paper presents a framework for recognizing multiple dump truck activities, i.e., idling, loading, hauling, and dumping, using Power Spectral Density (PSD) and time waveform images derived from vibration signals, along with a Convolutional Neural Network model. The PSD shows better evaluation metrics than the time waveform due to the presence of specific and distinct frequency components in PSD images for each dump truck activity. The results show that the accuracy of the activity recognition is not hindered by the number or position of accelerometers, i.e., even with a single accelerometer, very high accuracy for distinguishing various dump truck activities is achieved. This framework will reduce cycle time and operational delay, enhance productivity, and lower operational costs for the mining and construction industries, all at a minimal investment. • Framework for dump truck activity recognition using vibration signal and CNN model. • Utilization of PSD images from vibration signal for input to the AlexNet CNN model. • Automatic extraction of distinct features from PSD images associated with dump truck activities. • Measurement of vibration signal from dump truck operations within open-cast coal mines. • The method has been tested and evaluated using vibration data from two dump trucks. [ABSTRACT FROM AUTHOR]

Human activity recognition has increasingly received attention in recent years to track regular activities of people. Existing activity recognition approaches considerably contributed to the analysis of human behavior. However, they still confront numerous issues related to the variability of activities performed by people within dynamic environments. Generally, this variability renders the used training or ontology models with predefined activities unsuitable. Therefore, creating an activity recognition approach that is able to leverage dynamically new and unknown activities at runtime becomes important. In this paper, we propose a novel knowledge-driven activity recognition framework using smartphone. This framework envisions taking a knowledge-driven approach to reinforce the recognition accuracy and people's quality of life in the context of dynamic environments at runtime. More specifically, we propose an ontology-based context evolution along with a dynamic decision-making, so that new and unknown performed activities can be accurately recognized. Furthermore, we use a public activity recognition dataset to demonstrate the effectiveness of the proposed framework and show its advantage over a data-driven baselines in terms of accuracy. Experimental results reveal that our framework not only reinforces the accuracy, but also enables an effective activity learning when facing unknown activities at runtime. [ABSTRACT FROM AUTHOR]

DCR is a Distributed Collaborative Reasoning multi-agent model that aims to recognize human activities in smart homes from distributed, heterogeneous and dynamic sensor data. In this model, distributed agents with diverse classifiers, detect sensor stream data, make local predictions, communicate and collaborate to identify current activities. Conflict resolution strategies are applied to generate one final predicted activity when the local predicted activity of an agent is different from received predicted activities of other agents. In this paper, a possibilistic conflict resolution strategy, PCMCR , is proposed. Possibility theory is particularly efficient in combining multiple agents predictions that can be incomplete, uncertain, and conflicting. The PCMCR strategy deals with uncertainty factor which can be presented in the predictions of poor agents. It can take advantage of the complementary information given by each agent, even the weak ones. We experimentally test this strategy by performing an evaluation study on Aruba dataset. The obtained results indicate an enhancement in terms of accuracy, F-measure and G-mean metrics compared to the existing conflict resolution method max-trust of DCR, to the centralized model and to an existing distributed system. [ABSTRACT FROM AUTHOR]

Human posture estimation plays a significant role in the growth of intelligent nursing robot, a field that demands high accuracy and respect for privacy. Nevertheless, traditional approaches to enhancing data-driven studies in this domain often face challenges, primarily due to privacy concerns in sensitive healthcare environments. Federated Learning rises as the solution to the problem, as it not only improves the model learning ability but also protects the data privacy. In this paper, we proposed a Federated Learning Human Posture Recognition (FL-HPR) framework according to image and point cloud fusion. FL-HPR significantly enhances the information flow in the global model while ensuring the data privacy in local models, the benefits highlight the framework's potential in sensitive applications. Furthermore, the key innovation of FL-HPR is the optimization of the local dynamic graph edge convolution network of robot, which improves the recognition accuracy of individual body limb and enhance overall robustness. Experiments on Non-IID datasets illustrate that the presented FL-HPR remarkably outperforms non-federated learning methods, suggesting its potential to improve the accuracy of human joint estimation. The breakthrough indicates that the proposed FL-HPR can be integrated into the intelligent nursing robot in practical applications. For further explorations, the open-source code and videos are available at https://github.com/Hebut-LEVO/FL-HPR. • Proposing a multi-sensor fusion framework for nursing robots to recognize human body poses. • Introducing federated learning to posture recognition validates FL-HPR's Non-IID feasibility. • Adopting non-enveloping point cloud data, a model segments human body limbs in nursing scenarios. • Proposed method estimates human joints in border regions of adjacent body segments. [ABSTRACT FROM AUTHOR]

The world's population is rapidly aging, leading to an increase in the number of people who need care and a reduction in the number of potential workers who can give that care. Hence, in order to fight this worker shortage, scientific researchers proposed solutions, mainly prototypes, to maintain people at home. These solutions monitor the activities performed by people and can detect anomalies in people's behavior to assist them in a proper way and at the perfect time. In this paper, we propose a solution based on three ultra-wideband radars to recognize activities in a prototype apartment. More precisely, we processed the data provided by the radars with a conventional band-pass filter applied on each bin independently. Then, we extracted several features and performed dimensionality reduction with the help of the SelectKbest algorithm and the principal component analysis. Finally, we tested the proposed approach with Random Forest algorithm and the leave-one-subject-out strategy. The results obtained show an average improvement of approximately 13% accuracy compared to our previous work. [ABSTRACT FROM AUTHOR]

Ultra-wide band radar systems have gained even more popularity as an emerging technology in health-related research and particularly in the context of ambient intelligence for activity recognition inside smart environments. Indeed, recent technological advances have facilitated their implementation on smaller SoC at lower cost. However, observed practical behaviors not always meet theoretical prediction. Therefore, this paper presents an empirical evaluation of Novelda XeThru X4-based IR-UWB radars behavior. To this end, several experiments were conducted to provide insightful analyses to consider while exploiting such recent technology. We believe that these assessments will benefit to academic researchers by guiding their implementation choices to design new robust activity recognition systems. [ABSTRACT FROM AUTHOR]

Diagnosis of mobility disorder has become increasingly important in recent years. This paper investigates the feasibility of classifying six hand activities including two crucially Tremored and Non-Tremored motions, based on perturbed near-field radiation of the single body worn textile (BWT) antenna. The deployed antenna shows excellent biocompatibility and less vulnerability to electromagnetic absorption. Time-domain conversion of activities set using rational function approximation is investigated. Window cropping technique is exploited for data augmentation. Seven young and one elderly volunteer were used to assess the system's performance. The measurements were performed in a chamber with an anechoic ceiling and negligible electromagnetic emanation. The Discrete Wavelet Transform (DWT) is adopted to extract the time domain features based on the fluctuation of impedance along with sliding window. Support Vector Machine (SVM) is utilized for classification. Two more algorithms namely k-Nearest Neighbor (k-NN) and Naive Bayes (NB) also been applied to test the accuracy of the classification and it is found that SVM produces best outcome in every cases under investigation. The performance of the different participants for the wrist mount and shoulder mount are tested individually. The wrist mount performs quite well, except for the elderly subject, which obtains low accuracy in both scenarios but the shoulder wins by a margin of 0.3%. Clear distinction between the results of tremored and non-tremored activities is seen. The combined augmented data for wrist and shoulder mount is observed to be 2.1% and 1.2% higher than the non-augmented data set, respectively. [ABSTRACT FROM AUTHOR]

• This paper proposes a graph theory approach to perform the human activity recognition. • This paper proposes the correlation coefficient based method for generating the graph using the signals in the UCI-HAR dataset. • The predefined thresholds are used for determining whether the nodes are connected or not. • The features are updated according to the activities via the multi-aggregation fusion approach. • The random forest is used to classify these activities. This paper proposes a graph theory approach to perform the human activity recognition. However, as the most common signal employed for performing the activity recognitions is the motion signal while the motion signal is well structured, ordered and independent one another, the graph theory cannot be applied directly. To address this issue, this paper proposes the correlation coefficient based method for generating the graph using the signals in the UCI-HAR dataset. Here, the predefined thresholds are used for determining whether the nodes are connected or not. The features are updated according to the activities via the multi-aggregation fusion approach. Finally, the random forest is used to classify these activities. To demonstrate the effectiveness of our proposed method, the percentage accuracy and the macro averaged F1 score yielded by our proposed method with the graph weights are compared to those without the graph weights as well as with the multi-aggregator are compared with the mean aggregator. Also, our proposed method is compared to some common methods such as those based on the CNN and SVM. It is found that our proposed method can achieve the percentage accuracy up to 98.74%, which significantly outperforms the existing methods. [ABSTRACT FROM AUTHOR]

The goal of continuous smartphone authentication is to detect when the adversary has gained possession of the user's device post-login. This is achieved by triggering re-authentication at fixed, frequent intervals. However, these intervals do not take into account external information that might indicate that the impostor has gained physical access to the user's device. Continuous smartphone authentication typically relies on behavioral cues, such as hand movement and touchscreen swipes, that can be collected without interrupting the user's activity. Because these behavioral signals are characterized by relatively high error rates compared to physiological biometrics, their use at fixed intervals leads to unnecessary interruptions to the user's activity in case of a false reject, and to not recognizing the impostor in case of a false accept. To address these issues, in this paper we introduce a novel framework called SMARTCOPE: Smartphone Change Of Possession Evaluation. In this work, SMARTCOPE leverages smartphone movement signals collected during user activity to determine when the smartphone is no longer in the hands of its owner. When this occurs, SMARTCOPE triggers re-authentication. By using these signals, we are able to reduce the total number of re-authentication points while simultaneously lowering re-authentication error rates. Our analysis shows that our technique can reduce equal error rates by over 40%, from 7.8% to 4.6% using movement and keystroke features. Further, we show that SMARTCOPE can be used to transform a static (login-time) authentication system, such as face recognition, to a continuous re-authentication system, with a significant increase in security and limited impact on usability. [ABSTRACT FROM AUTHOR]

This paper proposes a framework to automatically determine the productivity and operational effectiveness of an excavator. The method estimates the excavator's actual, theoretical, and relative cycle times in the loading operation. Firstly, a supervised learning algorithm is proposed to recognize excavator activities using motion data obtained from four inertial measurement units (IMUs) installed on different moving parts of the machine. The classification algorithm is offline trained using a dataset collected via an excavator operated by two operators with different levels of competence in different operating conditions. Then, an approach is presented to estimate the cycle time based on the sequence of activities detected using the trained classification model. Since operating conditions can significantly influence the cycle time, the actual cycle time cannot solely reveal the machine's performance. Hence, a benchmark or reference is required to analyze the actual cycle time. In the second step, the theoretical cycle time of an excavator is automatically estimated based on the operating conditions, such as swing angle and digging depth. Furthermore, two schemes are presented to estimate the swing angle and digging depth based on the recognized excavator activities. In the third step, the relative cycle time is obtained by dividing the theoretical cycle time by the actual cycle time. Finally, the results of the method are demonstrated by the implementation on two case studies which are operated by inexperienced and experienced operators. The obtained relative cycle time can effectively monitor the performance of an excavator in loading operations. The proposed method can be highly beneficial for worksite managers to monitor the performance of each machine in worksites. • Excavator activities are recognized using inertial measurement units (IMUs). • Estimation of the actual cycle time of an excavator in the loading operation. • Estimation of the relative cycle time using the actual and theoretical cycle times. • Comparison of the performances of experienced and inexperienced operators. [ABSTRACT FROM AUTHOR]

Activity recognition plays a key role in many fields, such as health monitoring and elderly care. Handling changes in user habits is a significant technical challenge in activity recognition. Ideally, a model should adapt to newly emerging classes and concept drift dynamically. This paper proposes a novel semi-supervised class incremental learning method, namely, disagreement-based class incremental random forest (Di-CIRF). The proposed model can detect newly emerging classes and update a previously established activity recognition model through streaming data. First, it is necessary to identify novel candidates by employing the disagreement-based confidence voting mechanism and minimum bounding box (MBB)-based separation detection to annotate newly emerging data accurately. Then, the coarse coding-based cohesion detection strategy is adopted to filter out the true novelty instances. This paper also proposes the iterative MBB-based splitting strategy and the pseudo-instance generation mechanism in Di-CIRF for updating the activity model without retaining the trained data. According to experimental results on four public activity recognition datasets, Di-CIRF outperforms the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Intelligent systems are increasingly part of our everyday lives and have been integrated seamlessly to the point where it is difficult to imagine a world without them. Physical manifestations of those systems on the other hand, in the form of embodied agents or robots, have so far been used only for specific applications and are often limited to functional roles (e.g. in the industry, entertainment and military fields). Given the current growth and innovation in the research communities concerned with the topics of robot navigation, human–robot-interaction and human activity recognition, it seems like this might soon change. Robots are increasingly easy to obtain and use and the acceptance of them in general is growing. However, the design of a socially compliant robot that can function as a companion needs to take various areas of research into account. This paper is concerned with the navigation aspect of a socially-compliant robot and provides a survey of existing solutions for the relevant areas of research as well as an outlook on possible future directions. • Great demand for socially aware robot navigation, the topic is interdisciplinary. • Review of the research areas influencing socially aware robot navigation. • Active Vision, Robot Navigation, Action Recognition and HRI are crucial to the task. • This paper highlights influential tasks, challenges and datasets. • This paper discusses future research directions. [ABSTRACT FROM AUTHOR]

• We present "Deep Action Stamps (DeepActs)", a novel data representation which encodes actions in terms of spatial and motion information extracted from face, hands, and body. To the best of our knowledge, this is the first work which models the spatial and temporal dependencies between facial joints, hand joints, and body joints for action recognition. • We present DeepActsNet, an ensemble of Enhanced Convolutional Graph Networks (ECGN) that learn convolutional and structural features from different feature channels of Deep Actions Stamps. We also develop a lightweight strong baseline, which is more powerful than the previous methods in terms of recognition accuracy and computational efficiency. • We present ablation studies in terms of the benefits of combining spatial and motion information from face, hands, and body, and the significance of ensembling convolutional and structural features for improving accuracy of challenging action classes. Experiments on three public datasets show that our contributions consistently exceed the state-of-the-art performance on all datasets with considerable margins. Human action recognition from videos has gained substantial focus due to its wide applications in the field of video understanding. Most of the existing approaches extract human skeleton data from videos to encode actions because of the invariance nature of the skeleton information with respect to lightning conditions and background changes. Despite their success in achieving high recognition accuracy, methods based on limited body joints fail to capture the nuances of subtle body parts which are highly relevant for discriminating similar actions. In this paper, we overcome this limitation by presenting a holistic framework for combining spatial and motion features from the body, face, and hands to develop a novel data representation termed "Deep Actions Stamps (DeepActs)" for video-based action recognition. Compared to the skeleton sequences based on limited body joints, DeepActs encode more effective spatio-temporal features that provide robustness against pose estimation noises and improve action recognition accuracy. We also present "DeepActsNet", a deep learning based ensemble model which learns convolutional and structural features from Deep Action Stamps for highly accurate action recognition. Experiments on three challenging action recognition datasets (NTU60, NTU120, and SYSU) show that the proposed model produces significant improvements in the action recognition accuracy with less computational cost compared to the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

• The novel problem of finding spatio-temporal hard attention in skeleton videos using deep learning for human activity recognition is discovered. • A temporal agent and a special agent are proposed for respectively detecting informative frames and relevant joints in frames. • The two agents are trained using deep reinforcement learning. • A framework for cooperatively training the temporal and spatial agents is proposed. • The effectiveness of the proposed method in improving recognition performance and the training phase run time is demonstrated using three widely used bench-mark activity recognition datasets. [Display omitted] The use of skeleton data for activity recognition has become prevalent due to its advantages over RGB data. A skeleton video includes frames showing two- or three-dimensional coordinates of human body joints. For recognizing an activity, not all the video frames are informative, and only a few key frames can well represent an activity. Moreover, not all joints participate in every activity; i.e., the key joints may vary across frames and activities. In this paper, we propose a novel framework for finding temporal and spatial attentions in a cooperative manner for activity recognition. The proposed method, which is called STH-DRL, consists of a temporal agent and a spatial agent. The temporal agent is responsible for finding the key frames, i.e., temporal hard attention finding, and the spatial agent attempts to find the key joints, i.e., spatial hard attention finding. We formulate the search problems as Markov decision processes and train both agents through interacting with each other using deep reinforcement learning. Experimental results on three widely used activity recognition benchmark datasets demonstrate the effectiveness of our proposed method. [ABSTRACT FROM AUTHOR]

Currently, people's daily study, work, leisure are more and more concentrated in high-rise buildings. It is often difficult to quickly obtain the floor information when they are in a complex and densely populated indoor environment such as a commercial center or station. People's demand for indoor location services has already expanded from the original two-dimensional to three-dimensional. Smartphone is the best portable device for floor positioning and pedestrian navigation since it is embedded with Wi-Fi/Bluetooth/Cellular, inertial and barometer sensors. Smartphone-based floor positioning system utilizes the collected measurements from these embedded sensors to estimate pedestrian's floor information. Due to the high similarity of radio signals and the small air-pressure difference between adjacent floors, it is arduous to obtain high-precision floor information in multistory environments with low floor heights, staggered floor structures, stairwells or atrium structures. To address this issue, a lot of research has been carried out to obtain an accurate floor positioning by calculating the sensor data from the user's smartphone. However, there still a lack of a study that can comprehensively reflect and summarize the current state of these works. In this paper, we provide an elaborate review for the existing smartphone-based floor positioning techniques. It discusses their potential applications, current challenges and future research trends. The work not only advances a better understanding of indoor positioning technology, but possibly provides clear direction to the design and development of more accurate and robust floor positioning systems based on sensors embedded in smartphone. [ABSTRACT FROM AUTHOR]

Recently, vision-based methods have been widely used to analyze the construction productivity based on onsite videos owing to their low cost, simple deployment, and easy maintenance. However, existing vision-based methods rely on supervised learning for activity recognition, which is computationally intensive owing to the necessity of labeling large-scale training datasets. To address this problem, this paper describes a vision-based method for automatically analyzing excavators' productivities in earthmoving tasks by adopting zero-shot learning for activity recognition. The proposed method can identify activities of general construction machines (e.g., excavators and loaders) without pre-training or fine-tuning. To verify the feasibility, the proposed method has been tested on videos recorded from real construction sites. The accuracy values for activity recognition and productivity evaluation are 86% and 87.8%, respectively. • A vision-based method is proposed for productivity analysis in earthmoving. • Zero-shot learning method CLIP has been adopted for activity recognition of excavators. • Proposed activity recognition does not need pre-training and datasets. • The results achieved an accuracy of 86% for excavator activity recognition. • The results achieved an accuracy of 87.8% for productivity analysis. [ABSTRACT FROM AUTHOR]

Human activity recognition has become a pillar of today intelligent Human–Computer Interfaces as it typically provides more comfortable and ubiquitous interaction. This paper proposes a novel fuzzy-based deep learning-based algorithm to predict future sequences of activities from a given sequence of daily living activities of a subject wearing a lower limb exoskeleton. The engineering application concerns the challenging task of recognizing locomotion activities of the wearer in real-time, which is needed to ensure appropriate control of the robot during daily living activities. Indeed, real-time locomotion activity recognition is very challenging for controlling lower-limb exoskeletons. The model proposes a new adaptive kernel, based on the data features derived from the fuzzy rules on the input sequences to enrich the features of the activity sequences. Then, a CNN is applied to extract local subsequences from the whole sequences to identify local patterns in the convolution window. Finally, an attention-based GRU is incorporated into the model to extract meaningful parts of the time-series sequences. The results show high accuracy in the estimation of the transition between gait modes which is critical to ensure smooth control of the exoskeleton. The performance of the model is evaluated using the dynamic activity data gathered from different subjects. The proposed model outperforms the traditional models used in the literature. [ABSTRACT FROM AUTHOR]

• The first effort on the HMM state optimization problem, i.e., optimizing the number of states and the discriminative quality of the states itself. • New representation model for symbolic sequences using their transition probability distributions over the optimized HMM states called topics. • Formalization of the hierarchical model selection problem for topic learning with a novel topic-scatter criterion. • Learning the underlying topics by a newly defined HMM state clustering algorithm. • Experimental evaluation on human activity recognition and protein recognition with comparisons to the neural network-based auto-encoder. Sequence representation, which is aimed at embedding sequentially symbolic data in a real space, is a foundational task in sequence pattern recognition. It is a difficult problem due to the challenges entailed in learning the intrinsic structural features within sequences in small sample size cases, in an unsupervised way. In this paper, we propose to represent each symbolic sequence by its transition probability distribution over discriminating topics, formalized by a set of optimized Hidden Markov Model (HMM) states shared by all sequences. An efficient method, called Markovian state clustering with hierarchical model selection, is proposed to optimize the Markovian states and to adaptively determine the number of topics. The proposed method is experimentally evaluated on human activity recognition and protein recognition, and results obtained demonstrate its effectiveness and efficiency. [ABSTRACT FROM AUTHOR]

Recent researches on sensor based human activity recognition (HAR) are mostly devoted to designing various network architectures to enhance their feature representation capacity for raw sensor data. In this paper, we focus on strengthening the vanilla convolution without adjusting the model architectures in HAR scenario. Inspired by the idea of grouped convolution, we propose a novel heterogeneous convolution for activity recognition task, where all filters within a specific convolutional layer are separated into two uneven groups. Specifically, the sensor input is down-sampled into a low-dimensional embedding, which is then convolved by one filter group to recalibrate normal filters within the other group. The two filter groups can complement each other, which is very beneficial for augmenting the receptive field of sensor signals for HAR task. Extensive experiments are conducted on several benchmark HAR datasets, which consists of OPPORTUNITY, PAMAP2, UCI-HAR, USC-HAD as well as the Weakly Labeled HAR dataset. The results show that the baseline models can be significantly improved. Our heterogeneous convolution is simple and can easily be integrated into standard convolutional layers without increasing extra parameters and computational overhead. Finally, the actual operation of heterogeneous convolution is evaluated on an embedded Raspberry Pi platform. • The proposed approach can strengthen the basic convolution. • Our proposed method divides the filters into two groups but unevenly. • It can achieve SOTA performance without increasing computational burden. • The visualizing analysis of heterogeneous convolution is provided. • Actual operation is evaluated on a Raspberry Pi platform. [ABSTRACT FROM AUTHOR]

The development of smart wearable devices has driven the rapid progress of activity recognition. However, existing activity recognition methods are still struggling to recognize single arm swings due to coarse-grained sensor data segmentation. Refined arm-swing-wise data segmentation is vital in some specific cases, such as the rehabilitation of disabled patients. In this paper, we propose a smartwatch-based arm-swing-wise data segmentation approach for human activity recognition, which converts original sensor signals into square-wave signals to detect the cut-off points of each arm swing. Particularly, our method can adaptively adjust the window size and step size of a sliding window without considering the change of swing speed. Empirical evaluation on two datasets, a self-collected dataset and a publicly-available benchmark dataset, shows superior performance of our approach over other methods under different settings, such as classifiers, features, and wearing positions. [ABSTRACT FROM AUTHOR]

Compared with other probabilistic semantic learning algorithms, Markov Logic Network (MLN) learning can integrate existing knowledge fragments. However, the knowledge fragments increase as the MLN learning algorithm traverses candidate objects, making the execution time of the MLN learning algorithm long. To shorten the execution time, an MLN Cumulative Learning Algorithm (MCLA) that combines the MLN learning algorithm and cumulative learning is proposed. MCLA can unify the existing knowledge and the new–old knowledge, perform multi-task learning under appropriate circumstances, and directly call the existing knowledge to learn the new knowledge. This paper applies MCLA to single-resident indoor activity scenarios to verify the effectiveness of the learning algorithm. Experiments have proved that MCLA not only ensures the accuracy and improves the versatility of knowledge, but also dramatically decreases the learning time and effectively manages the knowledge memory of the MLN. • The increase of knowledge reserve in the Markov network reduces the MLN learning efficiency. • MCLA improves learning efficiency and knowledge storage by combining MLN learning and cumulative learning. • Creating a knowledge list K in MCLA improves the versatility of knowledge and enables multi-task learning. [ABSTRACT FROM AUTHOR]

• Propose an eigendecomposition-based method to select sensors in a high-efficiency way. • Propose a domain division-based method to update rules in a high-efficiency way. • Propose an online updating EBRBS for achieving activity recognition more efficiently. • Experiments show the proposed EBRBS has a high accuracy and excellent response time. Sensor-based activity recognition (AR) is a core problem with the research domain of smart environments. It has, however, the potential to provide solutions to address the problems associated with the growing size and ageing profile of the global population. The work presented within this paper focuses on the extended belief rule-based system (EBRBS), which offered promising performance compared with popular benchmark AR models and exhibited a high robustness in the situation of sensor failure. Nevertheless, efficiency remains one of the major issues to be improved for determining and updating the extended belief rule base (EBRB) within the EBRBS. This is critical for further utilizing the EBRBS in AR situations within dynamic smart environments. An eigendecomposition-based sensor selection method is firstly proposed to select an effective subset of sensors and to also enable efficient implementation to facilitate online AR. A novel domain division-based rule generation method is also proposed to generate and update an EBRB efficiently when new sensor data are available or when some sensors should be included or excluded in the EBRB. The combination of these two methods leads to an enhanced EBRBS, called online updating EBRBS. Two datasets (in a balanced class situation) obtained from simulation and actual environments are studied to provide detailed experimental analysis as a preliminary study and basis to handle further the imbalanced situation of real AR. The experimental results demonstrate an enhanced performance of the online updating EBRBS compared with the original EBRBS and some benchmark AR models, in terms of efficiency and effectiveness. [ABSTRACT FROM AUTHOR]