Dynamic Hand Gesture Recognition Using Electrical Impedance Tomography.

Hand gesture identification is one of problems being widely studied. There are two research trends corresponding to two data types, which are static and dynamic gestures. The static gesture is recognized based on the hand shape, while motion is the main feature in identifying dynamic gestures. In this paper, we propose an approach for modeling the dynamic hand gestures based on a combination of two mentioned information. At first, the hand silhouette is extracted using a skin-color filter. A sequence of geometric manipulations is then performed to remove the possible arm. The characteristics which describe the hand shape and motion orientation are estimated. Finally, the k-means clustering technique is combined with hidden Markov model to model each dynamic gesture. The experiments are performed on human-computer interaction dataset and obtain high efficiency.

This paper deals with static and dynamic hand gesture (digits) recognition. The method provides a threefold novel contribution: (1) segmentation algorithm gives better results on any skin colour and any size of hand on complex and non-uniform background; (2) key frame finding algorithm and (3) the recognition technique of signs (static digits, alphabets and dynamic digits). We separate out key frames from a sequence of static gestures, which include correct gestures from a video sequence. The recognition efficiency of key frame detection is 93% using the proposed algorithm. The segmentation efficiency is almost 95%. Features are extracted using the proposed feature extraction algorithm, and gestures are recognised. We propose a novel algorithm for static and dynamic gesture recognition. The proposed algorithm shows recognition efficiency of 94.8% for static gestures and 94% for dynamic gestures.

Sign language recognition plays a crucial role in supporting communication between deaf communities and hearing individuals. In particular, Vietnamese Sign Language (VSL) recognition remains a challenging task due to the complexity of hand gestures and limited available datasets. However, most existing approaches address static gestures and dynamic sign phrases as separate recognition problems, often employing different feature representations and independent processing pipelines. This fragmentation increases system complexity and limits scalability for real-time sign language applications. This study proposes a unified hand-landmark-based deep learning framework for recognizing both static and dynamic VSL gestures within a single integrated system. The system begins by detecting hand landmarks through the MediaPipe hand tracking pipeline. Based on temporal motion analysis of landmark sequences, a gesture routing mechanism automatically determines whether the input corresponds to a static gesture or a dynamic sign phrase. Static gestures are classified using a convolutional neural network (CNN), while dynamic gestures are processed using a long short-term memory (LSTM) network to capture temporal dependencies. Experiments were conducted on a VSL dataset consisting of 23 static hand signs and 4 dynamic sign phrases collected from multiple participants with variations in hand shape and gesture execution style. The system performance is evaluated using accuracy, precision, recall, and F1-score. Experimental results demonstrate that the proposed framework achieves an average recognition accuracy of 92% for static gestures and 88.4% for dynamic sign phrases, outperforming traditional machine learning baselines. The proposed system provides a practical and efficient solution for VSL recognition and has potential applications in real-time assistive communication systems.

Electrical brain imaging (EBI) refers to a set of techniques that exploit either the spontaneous electrical activity of the central nervous system, as in electroencephalographic (EEG) source reconstruction, or make use of external current injections, as in electrical impedance tomography (EIT) , to image the structure or function of the brain. When compared to other brain imaging methods used in research or in the clinical setting, such as computed tomography (CT), magnetic resonance imaging (MRI), functional MRI (fMRI), positron emission tomography (PET) and single photon emission computed tomography (SPECT), EIT and EEG source localization instrumentation offer the advantages of portability, low cost, high temporal resolution [ms] and quick setup. The downsides are a low spatial resolution [cm], high computational cost of the image reconstruction process and high sensitivity to imperfections of the electrical model of the head.
\nIn this work, a new special purpose reconstruction algorithm for EIT is presented and validated wth experimental measurements performed on a cylindrical phantom and on a simulated human head. The algorithm focuses on the quick detection of compact conductivity contrasts in imperfectly known in 3D domains. The performance of the proposed algorithm is then compared to the one of a benchmark reconstruction method in the EIT field, Tikhonov regularized reconstruction, with stroke detection and classification as a case study. Moreover, the possible application of EIT imaging to the detection of epileptic foci with intracranial deep electrodes (stereoelectroencephalography or SEEG) is explored.
\nFinally, EEG source reconstruction algorithms are implemented on a heterogeneous multi-CPU and multi-GPU computing system to significantly reduce the reconstruction time.

Electrical impedance tomography (EIT) is the process of spatially mapping the conductivity distribution of a domain based on voltage-current relationships observed at the boundary of the domain. Due to its low cost, ease of implementation, and benign nature (i.e. non-ionizing), EIT has received much attention for potential application to damage, strain, and saturation mapping in civil, mechanical, and aerospace venues that employ conductive materials (e.g. nanofiller-modified cements and polymers, carbon fiber composites, etc.). Mathematically, however, EIT is an ill-posed inverse problem predicated on reconstructing the conductivity distribution on a model of the domain. This process is prone to nonconvergence due to factors such as experiment-to-model domain shape discrepancies and electrode misplacement. Time difference imaging is commonly used to overcome this challenge–boundary measurements are collected when the material is in a pristine state and another set of boundary measurements are collected after some conductivity-changing perturbation such that experiment-to-model discrepancies largely subtract out. But this process of time-difference imaging is problematic for several reasons: one, a pristine state is not available for existent or already in-service structures. Two, pristine-state data can be lost over time. And three, the pristine-state data may no longer be applicable if there are significant shape changes, electrodes are reapplied, etc. Thus, it is desirable to have a baseline-free technique for EIT. To that end, we herein present the results of a preliminary investigation on damage detection via frequency-difference imaging (fdEIT). In this approach, we leverage the fact that materials generally exhibit frequency-dependent conductivity such that an EIT image can be formed from the difference between two different frequencies. Experimental measurements of AC conductivity up to 10 MHz were collected from a carbon nanofiber (CNF)-modified glass fiber/epoxy laminate. These measurements were then used in a computational simulation to explore the potential of fdEIT for damage detection and localization. The results of this study suggest that fdEIT may indeed be a viable method of eliminating baseline measurements for EIT thereby providing an important advance towards making EIT a practical inspection modality.

Individualized lung-protective ventilation strategy is essential for patients with COVID-19. In this study, a wireless, low-power, and miniaturized electrical impedance tomography (EIT) system was developed for remote and long-term monitoring of lung ventilation for patients with COVID-19 in the isolation ward of ICU. A new strategy of combining filtering and improved voltage-controlled current was employed to design the current source for a simplified system structure, and a new differential receiver circuit consisting of a buffer amplification and a differential amplification was proposed to design the voltage measurement unit for high precision. Moreover, a Bluetooth interface was adopted for wireless data transmission, and components characterized by low power consumption were selected to minimize the system power consumption. The proposed EIT system occupies a total size of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$6.5\times 4.5\times0.6$ </tex-math></inline-formula> cm <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup> and can work stably within 25 m from the wireless terminal installed with host EIT software. The system has a signal-to-noise of 70 dB at 50 kHz, stability of 0.1% relative change, and power consumption of 114 mW, as tested with a resistor phantom. Furthermore, a comparison between the proposed system and a commercial ICU EIT device was conducted on nine healthy volunteers for ventilation monitoring. The correlation between lung volume and relative impedance changes was higher than 0.9 for both EIT systems ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$p &lt; 0.001$ </tex-math></inline-formula> ). The analysis of the corresponding EIT images revealed that both systems delivered comparable images in terms of linearity, repeatability, and regional ventilation distribution. The test and experimental results suggested that the proposed system could conveniently provide reliable data acquisition (DAQ) for remote and long-term lung ventilation monitoring.

Rolling bearing is an essential part of various rotating machines, and its signal is the typical nonlinear signal. Traditional fault diagnosis usually relies on manual experience to extract the features of signals first. Deep convolutional neural networks (DCNN) can make fuller use of time series than traditional convolutional neural networks (CNN). Because of the low accuracy rate, fault diagnosis using gated recurrent unit (GRU) alone is not unsatisfactory. In order to improve the temporality of one-dimensional convolutional neural networks (1D-CNN) and enhance the accuracy of GRU, a novel fault diagnosis method called deep convolutional neural networks - gated recurrent unit (DCNN-GRU) is first put forward, which combines DCNN with GRU. The original signals without preprocessing are input into the DCNN, and the outputs of DCNN are input into the GRU consequently. Then the faults of rolling bearing can be diagnosed effectively. As the post-processing method, the t-distributed stochastic neighbor embedding (t-SNE) method is applied to visualize the fault diagnosis results. Six different network models, including the DCNN-GRU, are used to train the same fault dataset for comparison. The simulation results show that the proposed method can reach more than 99.9% accuracy stably for the given dataset, which can verify the feasibility and effectiveness of proposed method. And the DCNN-GRU can also be verified with good generalization ability using different dataset.

Due to the continuous growth and advancement of technology, an enormous volume of text data is generated daily across various sources including social media platforms, websites, search engines, healthcare records, and news articles. Extracting meaningful patterns from text data, such as viewpoints, related theories, journal distribution, facts, and the development of online news text, is a challenging task due to the varying lengths of the texts. One issue arises from the length of the text data itself, and another challenge lies in extracting valuable features, especially in news articles. In the deep learning models, the convolutional neural networks (CNNs) are capable of capturing local features in text data, but unable to capture the structural information or semantic relationships between words. Consequently, a sole CNN network often yields poor performance in text classification tasks, whereas the Gated Recurrent Unit (GRU) is adept at effectively extracting semantic information and understanding the global structural relationships present in textual data. This paper presents a solution to the problem by introducing a new text classification that integrates the strengths of CNN and GRU. The proposed hybrid models incorporate word vectorization and word dispersion in parallel. Initially, the model trains word vectors using the Word2vec model and then leverages the GRU model to capture semantic information from text sentences. Subsequently, the CNN method is employed to capture crucial semantic features, leading to classification using the SoftMax layer. Experimental findings demonstrated that the proposed hybrid GRU_CNN model outperformed and achieved accuracy 97.73% as compared to individual CNN, LSTM, and GRU models in terms of classification effectiveness and accuracy.

Electrical impedance tomography (EIT) is a medical imaging technique based on the injection of a current or voltage pattern through electrodes on the skin of the patient, and on the reconstruction of the internal conductivity distribution from the voltages collected by the electrodes. Compared to other imaging techniques, EIT shows significant advantages: it does not use ionizing radiation, is non-invasive and is characterized by high temporal resolution. Moreover, its low cost and high portability make it suitable for real-time, bedside monitoring. However, EIT is also characterized by some technical limitations that cause poor spatial resolution. The possibility to design wearable devices based on EIT has recently given a boost to this technology. In this paper we reviewed EIT physical principles, hardware design and major clinical applications, from the classical to a wearable setup. A wireless and wearable EIT system seems a promising frontier of this technology, as it can both facilitate making clinical measurements and open novel scenarios to EIT systems, such as home monitoring.

Electrical Impedance Tomography (EIT) is a technique that uses the voltage measurements, due to the injected alternating electrical currents to the body, taken on the boundary to produce an image that represents the internal conductivity distribution of the body. This is a relatively new imaging technique that has the potential to be used for clinical applications. As of today, EIT is not yet used in medical facilities as an alternative to X-ray and the likes, but it has the potential for clinical applications as it is a non-invasive and non-intrusive type of imaging technique. EIT is not that stable and as accurate compared to techniques that use radiation, since electrical current travels through paths with lesser impedance. The reconstruction of the image from the surface voltage measurements is difficult and challenging because it is nonlinear in nature. This study proposes (i) an innovative imaging technique that uses an inverse solver coupled with Convolutional Neural Network (CNN) trained on simulated data, (ii) and a difference EIT simulation system (Web.EIT) that features the proposed algorithm. The algorithm is evaluated and compared to an existing inverse solver (onestep Gauss-Newton) using simulated and actual datasets. Results showed that the proposed method outperformed the existing inverse solver by 37.02% lesser mean squared error (MSE) and 73.14% greater structural similarity (SSIM) index.

In recent years, channel state information (CSI) in WiFi 802.11n has been increasingly used to collect data pertaining to human activity. Such raw data are then used to enhance human activity recognition. Activities such as lying down, falling, walking, running, sitting down, and standing up can now be detected with the use of information collected through CSI. Human activity recognition has a multitude of applications, such as home monitoring of patients. Four deep learning models are presented in this paper, namely: a convolution neural network (CNN) with a Gated Recurrent Unit (GRU); a CNN with a GRU and attention; a CNN with a GRU and a second CNN, and a CNN with Long Short-Term Memory (LSTM) and a second CNN. Those models were trained to perform Human Activity Recognition (HAR) using CSI amplitude data collected by a CSI tool. Experiments conducted to test the efficacy of these models showed superior results compared with other recent approaches. This enhanced performance of our models may be attributable the ability of our models to make full use of available data and to extract all data features, including high dimensionality and time sequence. The highest average recognition accuracy reached by the proposed models was achieved by the CNN-GRU, and the CNN-GRU with attention models, standing at 99.31% and 99.16%, respectively. In addition, the performance of the models was evaluated for unseen CSI data by training our models using a random split-of-dataset method (70% training and 30% testing). Our models achieved impressive results with accuracies reaching 100% for nearly all activities. For the lying down activity, accuracy obtained from the CNN-GRU model stood at 99.46%; slightly higher than the 99.05% achieved by the CNN-GRU with attention model. This confirmed the robustness of our models against environmental changes.

Electrical impedance tomography with deep Calderón method

Chapter eight - Electrical impedance tomography image reconstruction based on autoencoders and extreme learning machines

Gesture recognition technology is widely used in the flexible and precise control of manipulators in the assisted medical field. Our MResLSTM algorithm can effectively perform dynamic gesture recognition. The result of surface EMG signal decoding is applied to the controller, which can improve the fluency of artificial hand control. Much current gesture recognition research using sEMG has focused on static gestures. In addition, the accuracy of recognition depends on the extraction and selection of features. However, Static gesture research cannot meet the requirements of natural human-computer interaction and dexterous control of manipulators. Therefore, a multi-stream residual network (MResLSTM) is proposed for dynamic hand movement recognition. This study aims to improve the accuracy and stability of dynamic gesture recognition. Simultaneously, it can also advance the research on the smooth control of the Manipulator. We combine the residual model and the convolutional short-term memory model into a unified framework. The architecture extracts spatiotemporal features from two aspects: global and deep, and combines feature fusion to retain essential information. The strategy of pointwise group convolution and channel shuffle is used to reduce the number of network calculations. A dataset is constructed containing six dynamic gestures for model training. The experimental results show that on the same recognition model, the gesture recognition effect of fusion of sEMG signal and acceleration signal is better than that of only using sEMG signal. The proposed approach obtains competitive performance on our dataset with the recognition accuracies of 93.52%, achieving state-of-the-art performance with 89.65% precision on the Ninapro DB1 dataset. Our bionic calculation method is applied to the controller, which can realize the continuity of human-computer interaction and the flexibility of manipulator control.

Low-pixel time-of-flight (ToF) sensors offer advantages in low power consumption, small size, and low cost. For gesture recognition, there are two advantages. Firstly, they are robust to changes in lighting conditions. Secondly, they could protect privacy. This paper addresses the challenge of utilizing an <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8}\times \mathbf{8}$</tex> pixel ToF sensor with extremely low resolution by proposing a signal processing system. The system includes pre-processing the data through non-global mean filtering to calibrate outliers. Additionally, we design a frames difference sliding window to ensure the integrity of gesture sequences and use Lagrangian interpolation to norm the fixed length. Finally, we design a tiny parallel convolutional network (TPC-Net) to recognize dynamic gestures. Experimental results demonstrate that the proposed system achieves an average accuracy of 98.47% over 6 dynamic gestures. Furthermore, our system achieves better performance with less model complexity compared to existing methods using low-pixel ToF sensors. These results illustrate the effectiveness of our system.