SUMMARYThe usage of depth data from time-of-flight or any equivalent devices overcomes visual challenges presented under low illumination. For integrating such data from multiple sources, the paper proposes a novel tool for the calibration of sensors. The proposed tool consists of retro-reflective stripped spheres. To correctly estimate these spheres, the paper investigates the performance of spherical estimations from either infrared or depth data. The relationship between sensors is determined by calculating the poses between the calibration tool and each sensor. The paper evaluates and compares the proposed approach against other state-of-the-art approaches in terms of shape reconstruction and spatial consistency.

This paper presents a framework of a marker-less human pose recognition system by identifying key body extremity parts through a network of calibrated low-cost depth sensors. The usage of depth sensors overcomes challenges related to low illuminations which usually compromises the information from the RGB cameras. Furthermore, the addition of multiple depth sensors complements the existing information with more visibility and less self-occlusion. A simple algorithm was applied which finds the connections between aligned and updated meshes produced from multiple sensors. These connections help to fuse the meshes into one large geodesic graph network. On this graph, a novel algorithm is applied to identify key body extremities such as head, hands, and feet of a human subject. A geodesic mapping is applied to the fused point cloud to produce a set of distinct topological clusters of 3D points. These clusters generate a hierarchical skeleton tree graph (Reeb graph) and produce a set of features for semantic identification of key body extremities. The combination of both the shape model and semantic classification finally leads to pose recognition. The paper presents the assessment of the proposed framework and its comparison with another available technique in a succession of experimental configurations.

This paper proposes a proof-of-concept, low-cost, and easily deployable Bluetooth low energy- (BLE-) based localization system which actively scans and localizes BLE beacons attached to mobile subjects in a room. Using the received signal strength (RSS) of a BLE signal and the uniqueness of BLE hardware addresses, mobile subjects can be identified and localized within the hospital room. The RSS measurement of the BLE signal from a wearable BLE beacon varies with distance to the wall-anchored BLE scanner. In order to understand and demonstrate the practicality of the relationship between RSS of a BLE beacon and the distance of a beacon from a scanner, the first part of the paper presents the analysis of the experiments conducted in a low-noise and nonreflective environment. Based on the analysis conducted in an ideal environment, the second half of the paper proposes a data-driven localization process for pinpointing the movements of the subject within the experimental room. In order to ensure higher accuracy like fingerprinting techniques and handle the increased number of BLE-anchored scanners like geometric techniques, the proposed algorithm was designed to combine the best aspects of these two techniques for better localization. The paper evaluates the effects of the number of BLE wall-mounted scanners and the number of packets on the performance of the proposed algorithm. The proposed algorithm locates the patient within the room with error less than 1.8 m. It also performs better than other classical approaches used in localization.

This paper explores tracking of specific body part by a network of multiple depth sensors. The usage of data from multiple depth sensors overcomes not only occlusion problems, but also visual challenges associated with RGB sensors under low illumination. Also, the identity of surveyed patient will be kept confidential as the silhouette of person is only shown in depth images. After determining relative poses between depth sensors, any point from depth image plane of any sensor can be transformed and projected onto depth image plane of another observing sensor. Such localization aid in overcoming occlusions observed in one sensor through obtaining information on the occluded parts from another sensor. Our paper introduces sensor scheduling system in which one of the multiple depth sensors' 3D co-ordinate frame is selected as world coordinate frame, contains fused point cloud and can track various limbs and parts such as head, hands and feet over time. To localize salient body extremities, a surface triangular mesh is applied on 3D fused point cloud with its corresponding generated geodesic extrema of that mesh coinciding with body extremities. The body extremities can be labelled based on those relative geodesic distances between body extremities. In evaluation, our multiple depth sensors system has managed to successfully localize specific body part i.e. head with less error against ground truth.

With growing demand for health care, medical clinicians and researchers need to gather quality patient centric data in order to better develop tools to improve quality of life. As an example, it has been a challenge to accurately evaluate postoperative mobility in patient care units. This paper describes the development and implementation of a proof-of-concept, a low cost and easily deployable Bluetooth Low Energy (BLE) based localization system used to identify and locate patients. Using the received signal strength (RSS) of BLE signal and the uniqueness of BLE hardware addresses, patients can be identified and located within the hospital room. The value of RSS of BLE signal from wearable BLE beacon varies with distance to the wall anchored BLE scanner. The first part of the paper presents the results of the experiments conducted in a low noise and non-reflective environment. The second half of the paper talks about the proposed algorithm of localization of the subject within the experimental room. The paper evaluates the effects of the number of BLE wall anchored scanners and the number of packets on the accuracy of the proposed algorithm. The proposed algorithm was evaluated to have better performance than other classical approaches.

This paper focuses on solving the problem of determining relative poses of the N cooperative sensors which include both mobile and stationary, along with position of targets. In the proposed approach, multi robots in a sensor network cooperate with each other in order to localize the position of each target and guide the robot to its assigned target position. The relative poses of neighboring sensors including target will be visually estimated and communicated with to other nearby sensors in the network. In addition, a network communication protocol was developed and implemented based on minimum hardware configuration. The localization and the tracking performance of our proposed approach was evaluated in our experimental hardware set-up.

This paper explores the novel use of multiple depth sensors to overcome occlusions and improve localization and tracking of body extremities. The usage of data from only depth sensors not only overcomes visual challenges associated with RGB sensors under low illumination, but also protects the identity of surveyed person with high confidentiality. For integrating depth information from multiple sources, the paper presents first an overview of a novel calibration method for multiple depth sensors. In case of occlusion of any fiducial point in the primary sensor's depth image, co-ordinates of the point can be obtained from the frame of other sensors using the calibration parameters. To localize salient body parts such as hands, head and feet, a surface triangular mesh is applied on generated 3D point cloud from the primary sensor. The geodesic extrema from the mesh coincide with body extremities. The body extremities can be identified based on those relative geodesic distances between the extremities. Once the body parts are labelled, a portion of body can be targeted and evaluated for specific gait analysis and visualization. For the performance evaluation, our calibration method has fared well in comparison to other available techniques. Also, our proposed localization of salient body parts is able to successfully tag the specific body part i.e. the head region.

Several efforts are being made in studying sleeping posture of a person under natural conditions without causing discomfort. In this study, various methods are proposed and explored in which features defining sleep postures of a person are investigated for intelligent pattern recognition. Using the measured depth data, three dimensional depth scans as well as the cross-sectional scans of the static sleeping body are captured. 1D Fast Fourier Transforms are used as complementary feature of the cross-sectional scans in pattern recognition. Due to better orientation-frequency characteristics, the two dimensional Gabor filter banks are also applied on depth images to produce Gabor images from which features known as Gabor features can be extracted. These selected features can be further integrated as a part of natural sleep posture recognition.

This paper presents experimental results from a platform consisting of multiple wearable body area networks (BAN) connected to a wireless mesh network. The proposed platform collects, processes and wirelessly transmits medical data from multiple wearable BAN to a medical control center (MCC) through a solar-powered and multi-hop mesh network. This proof-of-concept platform encompasses several innovations. In the BAN, a dynamic TDMA MAC layer has been implemented over a 802.15.4 physical layer as well as 2 lightweight protocols. To reduce the number of packet sent by sensors and the size of packet sent by the gateway, a similarity-based filter and a polynomial interpolation technique respectively are used. In the solar-powered mesh network, a machine learning algorithm has been implemented to predict battery depletion and ensure continuity of service.

To gather and transmit data, low cost wireless devices are often deployed in open, unattended and possibly hostile environment, making them particularly vulnerable to physical attacks. Resilience is needed to mitigate such inherent vulnerabilities and risks related to security and reliability. In this paper, Routing Protocol for Low-Power and Lossy Networks (RPL) is studied in presence of packet dropping malicious compromised nodes. Random behavior and data replication have been introduced to RPL to enhance its resilience against such insider attacks. The classical RPL and its resilient variants have been analyzed through Cooja simulations and hardware emulation. Resilient techniques introduced to RPL have enhanced significantly the resilience against attacks providing route diversification to exploit the redundant topology created by wireless communications. In particular, the proposed resilient RPL exhibits better performance in terms of delivery ratio (up to 40%), fairness and connectivity while staying energy efficient.

This paper presents experimental results on a fully functional body area network (BAN) platform in terms of energy consumption, delivery ratio and lifespan. The proposed BAN platform captures, processes, and wirelessly transmits six-degrees-of-freedom inertial and electrocardiogram data in a wearable, non-invasive form factor. A dynamic TDMA MAC layer has been implemented over a 802.15.4 physical layer as well as 2 lightweight protocols: a similarity-based filter to reduce the number of packets sent by the sensors and a polynomial interpolation technique to reduce the size of the packets sent by the coordinator. The system is evaluated regarding the delivery rate, the energy consumption efficiency and the lifetime while considering 3 scenarios and several human activities (sitting, walking, and running). In addition, we compare the sensors' lifespan when bluetooth or 802.15.4 is used. The experimental results show that the proposed MAC layer reduces the number of collisions and is particularly adapted to periodic data traffic from biomedical sensors. Moreover, significant improvements in energy consumption and lifetime are observed enabling health care applications and remote monitoring in harsh environments.

This paper presents a 12-lead electrocardiogram (ECG) wireless medical system (WMS) aimed to provide continuous patient monitoring. The WMS consists of three sections: end device (ED), base station (BS) and graphical user interface (GUI). The ED and BS communicate through an RF link that operates in the industrial, scientific and medical (ISM) band at 2.45 GHz. The ED is based on commercially-available ultra-low power integrated circuits, including an 8 channel analog front-end (AFE) intended for biopotential monitoring, a microcontroller unit and an ISM radio transceiver. Details on the 12-lead ECG WMS operation including the low-power oriented control algorithm of the ED and the detection algorithm embedded in the GUI are presented. Measurements of the WMS using a 10 electrode ECG signal generator show successful communication up to 35m line of sight. The ED average current consumption is 33mA (including RF transmission) in continuous operation. This allows the ED to operate up to 88 hours.

This paper investigates the tradeoff between accuracy and complexity cost to predict electrocardiogram values using auto-regressive moving average (ARMA) models in a fully functional body area network (BAN) platform. The proposed BAN platform captures, processes, and wirelessly transmits six-degrees-of-freedom inertial and electrocardiogram data in a wearable, non-invasive form factor. To reduce the number of packets sent, ARMA models are used to predict electrocardiogram (ECG) values. However, in the context of wearable devices, where the computing and memory capabilities are limited, the prediction model should be both accurate and lightweight. To this end, the goodness of the ARMA parameters is quantified considering ECG signal, we compute Akaike Information Criterion (AIC) on more than 900000 ECG measures. Finally, a tradeoff is given accordingly to the hardware constraints.

To gather and transmit data, low cost wireless devices are often deployed in open, unattended and possibly hostile environment making them particularly vulnerable to physical attacks. Resilience is needed to mitigate such inherent vulnerabilities and risks related to security and reliability. In this work, Routing Protocol for Low-Power and Lossy Networks (RPL) is studied in presence of packet dropping malicious compromised nodes. Random behavior and data replication have been introduced to RPL to enhance its resilience against such insider attacks. The classical RPL and its resilient variants have been analyzed through simulations. Resilient techniques introduced to RPL have enhanced significantly the resilience against attacks providing route diversification to exploit the redundant topology created by wireless communications. In particular, the proposed resilient RPL exhibits better performance in terms of delivery ratio (up to 40%), fairness and connectivity while staying energy efficient.

This paper presents a comprehensive approach for the detailed analysis of ECG waveforms including various morphologies to aid clinical diagnosis. Clinical judgment is often based on observing various features which may occur simultaneously on the ECG. Thus, to automate diagnosis, a comprehensive tool capable of detecting all these features is required. Parabolic curve fitting, adaptive thresholds and synchronicity across leads are utilized to detect the various waves of the QRS complex namely Q,R,S,R’ and S’. Onset of the QRS complex and the J point are detected using a ‘modified second derivative’ approach. The isoelectric level is detected using linearity and slope conditions. P and T waves are detected using ‘area under curve’ approach. Measurements such as peak-to-peak intervals and ST elevation/depression are numerically calculated from the points obtained. Curve fitting and change in slope are utilized for obtaining morphology of the ST segment. Presence of significant Q waves and abnormal T waves are inferred using clinical guidelines and numerical calculations. The performance of the algorithm is validated on 40 sample patient data — 20 healthy and 20 with Myocardial Infarction. Average accuracy shown in detecting all points of interest is 98.5%. All measurements are successfully calculated from these points. Along with this reliable performance, the approach proves to be simple and computationally fast.

This work presents the development of an algorithm for analyzing ECG waveforms. The identification of the various waveforms on an ECG is the first and most crucial step in any automated analysis. The algorithm developed is capable of detecting all important waveforms. These include the Q,R,S,R' and S' waves, J and ST points and onset and offset of P and T waves. The various techniques utilized in their detection include adaptive thresholding, parabolic curve fitting, modified derivatives and temporal coherence. After detecting the waveforms, various measurements are obtained from the points detected. These include the polarity of the waveforms, the ST elevation, morphology of ST segment, QRS width and T wave morphologies. These measurements can easily be utilized for diagnosing the ECG for various abnormalities. The developed detection and measurement algorithm stands out from previous works in a number of ways. It is comprehensive and capable of detecting all the important ECG waveforms. It works well even on atypical ECG beats with secondary R' and S' waves. The adaptive thresholding approach minimizes the dependence on fixed hard thresholds. The synchronous behavior of ECG recordings across leads is exploited to improve accuracy. The performance of the algorithm was validated on 40 sample 12-lead ECG data of two minutes each. To validate performance on abnormal ECG waveforms, 20 datasets from patients diagnosed for myocardial infarction (MI) were included. An overall detection accuracy of 98.5% was obtained.