This paper presents a multimodal system for seamless surveillance of elderly people in their living environment. The system uses simultaneously a wearable sensor network for each individual and premise-embedded sensors specific for each environment. The paper demonstrates the benefits of using complementary information from two types of mobility sensors: visual flow-based image analysis and an accelerometer-based wearable network. The paper provides results for indoor recognition of several elementary poses and outdoor recognition of complex movements. Instead of complete system description, particular attention was drawn to a polar histogram-based method of visual pose recognition, complementary use and synchronization of the data from wearable and premise-embedded networks and an automatic danger detection algorithm driven by two premise- and subject-related databases. The novelty of our approach also consists in feeding the databases with real-life recordings from the subject, and in using the dynamic time-warping algorithm for measurements of distance between actions represented as elementary poses in behavioral records. The main results of testing our method include: 95.5% accuracy of elementary pose recognition by the video system, 96.7% accuracy of elementary pose recognition by the accelerometer-based system, 98.9% accuracy of elementary pose recognition by the combined accelerometer and video-based system, and 80% accuracy of complex outdoor activity recognition by the accelerometer-based wearable system.
The aim of this paper is to choose the optimal motion sensor for the selected human activity recognition. In the described studies, different human motion measurement methods are used simultaneously such as optoelectronics, video, electromyographic, accelerometric, and pressure sensors. Several analyses of activity recognition were performed: recognition correctness for all activities together, matrices of the recognition errors of the individual activities for all volunteers for the individual sensors, and recognition correctness of all activities for each volunteer and each sensor. The experiments enabled to find a range of interchangeability and to choose the most appropriate sensor for recognition of the selected motion.
The design of artificial limbs is a research topic that has, over time, attracted considerable interest from researchers in various fields of study, such as mechanics, electronics, robotics, and neuroscience. Continuous efforts are being made to build electromechanical systems functionally equivalent to the original limbs and to develop strategies to control them appropriately according to the intentions of the user. The development of Human–Machine Interfaces (HMIs) is a key point in the development of upper limb prostheses, since the actions carried out with the upper limbs lack fixed patterns, in contrast to the more predictable nature of lower limb movements. This paper presents the development of an HMI system for the control of a transhumeral prosthesis. The HMI is based on a hybrid control strategy that uses voice commands to trigger prosthesis movements and regulates the applied grip strength when the user turns his head. A prototype prosthesis was built using 3D technology and trials were conducted to test the proposed control strategy under laboratory conditions. Numerical simulations were also performed to estimate the grip strength generated. The results obtained show that the proposed prosthesis with the dedicated HMI is a promising low-cost alternative to the current solutions. The proposed hybrid control system is capable of recognizing the user’s voice with an accuracy of up to 90%, controlling the prosthesis joints and adjusting the grip strength according to the user’s wishes.
1 March 2020 in Journal of Physics: Conference Series
This paper proposes a robust biometric identification system using compressed electrocardiogram (ECG) signal by varying physiological conditions. The ECG data were obtained by recording a total of 30 healthy subjects where they performed six regular daily activities repeatedly at a sampling frequency of 1000 Hz. Then, the QRS complexes are segmented by implementing Amplitude Based Technique (ABT) where it compares the amplitudes of ECG points to determine the R peak. The segmented QRS is then compressed for various levels by using Discrete Wavelet Transform (DWT) algorithms and first 3 Daubechies (db) wavelet are computed. Next, a Cardioid graph is generated. In order to verify the matching process, the classification is performed by using the Multilayer Perceptron (MLP) technique. The results show that by applying this method, the accuracy of the identification rate can be achieved as high as 96.4% even when the data file is compressed up to 73.3%. When the data file is compressed, the outcomes also demonstrate that the execution time is less compare to non-compressed data. Therefore, the biometric identification system can be implemented efficiently as there will be a lesser issue regarding the data storage, execution time and accuracy based on the outcome of the study.
Multimodal sensing and data processing have become a common approach in modern assisted living systems. This is widely justified by the complementary properties of sensors based on different sensing paradigms. However, all previous proposals assume data fusion to be made based on fixed criteria. We proved that particular sensors show different performance depending on the subject’s activity and consequently present the concept of an adaptive sensor’s contribution. In the proposed prototype architecture, the sensor information is first unified and then modulated to prefer the most reliable sensors. We also take into consideration the dynamics of the subject’s behavior and propose two algorithms for the adaptation of sensors’ contribution, and discuss their advantages and limitations based on case studies.
This study presents a comparative analysis of electrocardiogram (ECG) and accelerometer (ACC) data in the context of unsupervised human activity recognition and subject identification. Recordings were obtained from 30 participants performing activities of daily living such as walking, sitting, lying, cleaning the floor, and climbing stairs. Distance-based signal comparison methods and clustering techniques were employed to evaluate the feasibility of each modality, both individually and in combination, to discriminate between individuals and activities. Results indicate that ACC signals provide superior performance in activity recognition (NMI = 0.728, accuracy = 0.817), while ECG signals show higher discriminative power for subject identification (NMI = 0.641, accuracy = 0.500). In contrast, combining ACC and ECG signals yielded lower scores in both tasks, suggesting that multimodal fusion introduced additional variability. These findings highlight the importance of selecting the most appropriate modality depending on the recognition objective and emphasize the challenges associated with multimodal approaches in unsupervised scenarios.