Detection Of Movement Patterns Research Articles

Human Activity Recognition Based on CNN and LSTM

<p>Human activity recognition (HAR) based on wearable devices is an emerging field of great interest. HAR can provide additional information on a human subject’s physical status. Utilising new technologies for HAR will become very meaningful with the development of deep learning. This study aims to mine deep learning models for HAR prediction with the highest accuracy on the basis of time-series data collected by mobile wearable devices. To this end, convolutional neural networks (CNN) and long short-term memory neural networks (LSTM) are combined in a deep network model to extract behavioural facts. The proposed CNN model contains two convolutional layers and a maximum pooling layer, and batch normalisation is added after each convolutional layer to improve convergence speed and avoid overfitting. This structure yields significant results in terms of performance. The model is evaluated on the MHEALTH dataset with a test set accuracy of 99.61% and can be used for the intelligent recognition of human activity. The results of this study show that the proposed model has better robustness and motion pattern detection capability compared to other models.</p> <p> </p>

Implementation of a sensor node for monitoring and classification of physiological signals in an edge computing system

We describe the design and development of sensor nodes, based on Edge computing technologies, for the processing and classification of events detected in physiological signals such as the electrocardiographic signal (ECG is the electrical signal of the heart), temperature, heart rate, and human movement. The edge device uses a 32-bit Tensilica microcontroller-based module with the ability to transmit data wirelessly using Wi-Fi. In addition, algorithms for classification and detection of movement patterns were implemented to be implemented in devices with limited resources and not only in high-performance computers. The Internet of Things and its application in smart environments can help non-intrusive monitoring of daily activities by implementing support vector machine (SVM is a machine learning algorithm) for implementation in embedded systems with low hardware resources. This paper shows experimental results obtained during the acquisition, transmission, and processing of physiological signals in a edge computing system and their visualization in a web application.

A Novel Compensatory Motion Detection Method Using Multiple Signals and Machine Learning

Harmful compensatory motion will seriously influence the upper limb rehabilitation effect of stroke patients. Existing compensatory motion pattern detection methods, which are mainly based on accelerometers, inertial measurement units (IMUs), and vision, have problems such as high installation requirements, calibration difficulty, object occlusion, and complex operations. To solve these problems, by analyzing the principle of compensatory motion, this paper designs a trunk restraint belt, compensation motion mode detection system that is composed of a force sensor (F-S), angular displacement sensor (AD-S), and surface electromyography (sEMG). Fifteen healthy subjects without dyskinesia completed normal motion and simulated compensatory movement modes (trunk rotation, trunk forward tilt, and scapular elevation) based on the system. Machine learning (ML) methods are used to classify three compensated motion and normal motion (NM). Among them, the support vector machine (SVM) classifier shows good classification performance for trunk forward tilt (F1-score = 0.9256), shoulder blade elevation (F1-score = 0.9190), trunk rotation (F1-score = 0.8262) and normal motion (F1-score = 0.9758). This method is a new compensation detection method with obvious advantages: it is easy to wear, and without considering the problems of calibration and target occlusion, it can not only identify the patient motion mode but also limit the large range of compensatory motion.

Visual Leakage Inspection in Chemical Process Plants Using Thermographic Videos and Motion Pattern Detection.

Liquid leakage from pipelines is a critical issue in large-scale chemical process plants since it can affect the normal operation of the plant and pose unsafe and hazardous situations. Therefore, leakage detection in the early stages can prevent serious damage. Developing a vision-based inspection system by means of IR imaging can be a promising approach for accurate leakage detection. IR cameras can capture the effect of leaking drops if they have higher (or lower) temperature than their surroundings. Since the leaking drops can be observed in an IR video as a repetitive phenomenon with specific patterns, motion pattern detection methods can be utilized for leakage detection. In this paper, an approach based on the Kalman filter is proposed to track the motion of leaking drops and differentiate them from noise. The motion patterns are learned from the training data and applied to the test data to evaluate the accuracy of the method. For this purpose, a laboratory demonstrator plant is assembled to simulate the leakages from pipelines, and to generate training and test videos. The results show that the proposed method can detect the leaking drops by tracking them based on obtained motion patterns. Furthermore, the possibilities and conditions for applying the proposed method in a real industrial chemical plant are discussed at the end.

FingerAuth: 3D magnetic finger motion pattern based implicit authentication for mobile devices

Smart devices, as the most widely used platforms for the mobile cyber–physical system (CPS) applications, such as smart home and health care systems, are becoming the prime targets of various attackers for users’ considerable private and confidential data in them. To fight against side channel attacks aiming to obtain credentials, e.g., passwords, during the process of user authentication, touch pattern based implicit authentication has been proposed. However, such a defensive technique fails to obtain an entire pattern of user operation by deriving user operation data via a touch-enabled screen. Considering that user operations, including on-screen and in-air finger movements, are performed in three-dimensional (3D) space, we propose a novel 3D magnetic finger motion pattern based implicit authentication technique, referred to as FingerAuth. To use FingerAuth, a user operates on her mobile device, e.g., texting a message and browsing websites, with a magnetic ring on the finger she uses. With the help of a built-in three-axis magnetometer on the mobile device, we can derive the 3D magnetic finger motion pattern as a human behavioral feature for implicitly authenticating the user. By using machine learning techniques, a robust 3D magnetic finger motion pattern detection model can be constructed. Two rounds of usability tests are conducted for the evaluation of FingerAuth. In the initial usability test targeting a given group of smart device users, we test the uniqueness of the proposed trait in typing scenario, achieving high average accuracy of 96.38%, low average false acceptance rate (FAR) of 4.06%, and false rejection rate (FRR) of 3.18%. In the second user usability test, we further evaluate the permanence of 3D finger motion pattern in multiple user–device interaction scenarios. There is an interim of two-week period between the training data collection phase and the testing data collection phase. The results of the high accuracy of over 80%, as well as the FAR and FRR of below 15%, indicate the applicability of FingerAuth.

Solving the challenges of monitoring mobile populations: insights from studies of waterbirds in southern Africa

Highly mobile species are difficult to monitor. It is often unclear from abundance estimates whether a population is genuinely declining or simply elsewhere. If the spatial and temporal scales of monitoring are inadequate, the data will only be useful if the signal is very strong (i.e. the change in population size is large). We explored the ability of three major databases – the Southern African Bird Atlas Project (SABAP), the Coordinated Waterbird Counts (CWAC) and the South African Bird Ringing Unit (SAFRING) – to reveal the movement patterns of a highly mobile African waterbird, the Red-billed Teal Anas erythrorhyncha. Only ring recovery data provided clear evidence of long-distance movement. All three data sets were compromised to varying degrees by methodological biases and spatial and/or temporal resolutions that prevented the detection of movement patterns. By incorporating abundance estimates, increasing and quantifying sampling effort, and extending coverage at the data collection stage, the potential for these data sets to contribute to our understanding of bird movements could be greatly improved. Monitoring programmes for mobile species must be tested at an early stage to assess their ability to explain spatial and temporal population variation at the scales at which such variation occurs.

Dynamic facial expression analysis based on extended spatio-temporal histogram of oriented gradients

Facial expression is crucial for proper analysis of a person's face. It is an indicator of the emotion of a person and thus has attracted the attention of many researchers. In this work, a novel local spatio-temporal descriptor is proposed for motion pattern detection. The proposed feature comprises histogram of 3D gradients and the gradients' variation over time to robustly describe the spatial and temporal information. It also incorporates spatio-temporal pyramid structure to handle different resolution and frame rate. To reduce the dimension of the feature, we applied genetic algorithm for region-based feature selection. We evaluated the performance of our proposed descriptors on facial expression recognition using the Cohn-Kanade CK

Similar adaptation effects on motion pattern detection and position discrimination tasks: Unusual properties of global and local level motion adaptation

Here we examine adaptation effects on pattern detection and position discrimination tasks in radial and rotational motion patterns, induced by adapting stimuli moving in the same or opposite directions to the test stimuli. Adaptation effects on the two tasks were similar, suggesting these tasks are performed by the same population of neurons. Global motion specific adaptation was then induced by presenting adaptation stimuli and test stimuli in different parts of the visual field. Again, adaptation effects on the two tasks were similar, but neither same-direction nor opposite-direction motion produced any adaptation effect on contracting motion patterns. Finally, adaptation stimuli were compared that should have similar effects on local motion processing neurons, but different effects on global motion processing neurons. Again, adaptation effects on the two tasks were similar. However, when global-level adaptation was avoided, no adaptation effects were seen with adaptation patterns moving in the opposite direction to the test pattern. Together, these last two experiments suggest that adaptation to opposite directions of motion from the test motion affects global motion processing but not local motion processing neurons.

Detecting motion patterns via direction maps with application to surveillance

Detection of motion patterns in video data can be significantly simplified by abstracting away from pixel intensity values towards representations that explicitly and compactly capture movement across space and time. A novel representation that captures the spatiotemporal distributions of motion across regions of interest, called the “Direction Map,” abstracts video data by assigning a two-dimensional vector, representative of local direction of motion, to quantized regions in space-time. Methods are presented for recovering direction maps from video, constructing direction map templates (defining target motion patterns of interest) and comparing templates to newly acquired video (for pattern detection and localization). These methods have been successfully implemented and tested (with real-time considerations) on over 6300 frames across seven surveillance/traffic videos, detecting potential targets of interest as they traverse the scene in specific ways. Results show an overall recognition rate of approximately 91% hits vs 8% false positives.