AniGaussian: Animatable Gaussian Avatar With Pose‐Guided Deformation

We present an upper-body exoskeleton for rehabilitation, called Harmony, that provides natural coordinated motions on the shoulder with a wide range of motion, and force and impedance controllability. The exoskeleton consists of an anatomical shoulder mechanism with five active degrees of freedom, and one degree of freedom elbow and wrist mechanisms powered by series elastic actuators. The dynamic model of the exoskeleton is formulated using a recursive Newton–Euler algorithm with spatial dynamics representation. A baseline control algorithm is developed to achieve dynamic transparency and scapulohumeral rhythm assistance, and the coupled stability of the robot–human system at the baseline control is investigated. Experiments were conducted to evaluate the kinematic and dynamic characteristics of the exoskeleton. The results show that the exoskeleton exhibits good kinematic compatibility to the human body with a wide range of motion and performs task-space force and impedance control behaviors reliably.

In soccer matches, 3D reconstruction is a main part of many applications including free-viewpoint broadcasting, match analysis, augmented reality, and refereeing examination. The main challenge of 3D reconstruction in soccer scenes is the human body reconstruction. Although 3D reconstruction methods have been improved to a high extent in controlled condition, still there are lots of uncovered issues in outdoor scene reconstructions with uncontrolled conditions. In this paper, a view-dependent depth optimization method is presented which addresses two of these issues in soccer scenes. These issues include inaccurate camera calibrations and limited number of input cameras. This method will optimize the proposed energy functions to find the optimal depth image related to each input camera. For computing the initial depth images, a visual hull computation technique is used. An energy function based on depth images is defined which includes four energy terms. Two new energy terms including silhouette term and pose term are introduced in this function. The energy function is optimized with the α-expansion algorithm to find the optimal depth image. This optimization is done for each camera, separately. Finally, the 3D model is reconstructed with the resulted depth images. Experimental results prove that compared to previous methods, better quality 3D models are achieved. Also, as an application, free-viewpoint videos are created. The experimental results show the efficiency of the proposed method.

Detection and isolation of battery charging cyberattacks via Koopman operator

Model-based evaluation of clustering validation measures

In the present work, a simple algorithm for stabilizing an unknown linear time-invariant system is proposed, assuming only that this system is stabilizable. The suggested algorithm is based on first performing a partial identification of the system and then stabilizing the controllable subsystem. It should be emphasized that our approach does not depend on any prior model knowledge and requires only a minimal number of samples.

Many target localization techniques based on received signal strength indicator (RSSI) assume a prior knowledge of a channel model and its parameters for an area where a target node is localized. This is a limiting factor since an intensive pre- measurement campaign is required to determine the model and its parameters. Basing on channel measurement campaigns in different areas we confirmed that the variation in the RSSIs of the IEEE 802.15.4 signal can be described by a two-layer model. We therefore propose a novel target localization method with no prior knowledge of model parameter values. The method is based on a joint maximum likelihood target localization and channel model parameters estimation. The proposed method was experimentally verified in real environments and the results show that the location estimation accuracy outperforms the accuracy of the conventional method where the values of the channel model parameters are known in advance.

Towards robust deep learning-based autosegmentation in MRI-planned gynecological brachytherapy: Importance of scalable development and comprehensive evaluation.

In this paper, a generative adversarial network based on SMPL model is used to reconstruct three-dimensional (3D) human mesh from a single image. The generative adversarial network consists of two parts (generative network and adversarial network). The generative network is made up of a deep residual network and a multilayer perception network. The goal is to generate suitable SMPL model parameters and camera parameters which fit the 2D and 3D position annotation of human skeleton joints. The adversarial network judges whether the shape and pose parameters generated by the generative network match the real human body through a discriminator. The purpose is to guide the generative network output parameters that conform to the laws of human kinematics. A geometric prior of the 3D joint angle limitation of the human body is introduced into the network to further constrain the generated human pose. Experiments verify the effectiveness of the algorithm.

The advances and availability of technologies for the acquisition, registration and analysis of the three-dimensional (3D) shape of human bodies (or body parts) are resulting in the formation of large databases of parameterised meshes from which digital human body models can be derived. Such models can be used for the data-driven reconstruction of parameterised human body shapes from partial information such as one-dimensional (1D) measurements or 2D images. In this paper, we propose a new method for the reconstruction of 3D bodies from images gathered with a smartphone or tablet. Moreover, the method is implemented into a prototype app and tested at different levels through three experimental studies including synthetic models, 1:10 scale figurines and real children. The results demonstrate the feasibility of acquiring reliable anthropometric information easily at home by non-experts. This method and implementation have great potential for their application to the personalisation, size recommendation and virtual try-on simulation of wearable products.

The objective of any multimodal medical image fusion algorithm is to assist a radiologist for better decision-making during the diagnosis and therapy by integrating the anatomical (magnetic resonance imaging) and functional (positron emission tomography/ single-photon emission computed tomography) information. We proposed a new medical image fusion method based on content-based decomposition, Principal Component Analysis (PCA), and sigmoid function. We considered Empirical Wavelet Transform (EWT) for content-based decomposition purposes since it can preserve crucial medical image information such as edges and corners. PCA is used to obtain initial weights corresponding to each detail layer. In our experiments, we found that direct usage of PCA for detail layer fusion introduces severe artifacts into the fused image due to weight scaling issues. In order to tackle this, we considered using the sigmoid function for better weight scaling. We considered 24 pairs of MRI-PET and 24 pairs of MRI-SPECT images for fusion, and the results are measured using four significant quantitative metrics. Finally, we compared our proposed method with other state-of-the-art transformbased fusion approaches, using traditional and recent performance measures. An appreciable improvement is observed in both qualitative and quantitative results compared to other fusion methods.

With the increasing utilization of X-ray Computed Tomography (CT) in medical diagnosis, obtaining higher quality image with lower exposure to radiation is a highly challenging task in image processing. Sparse representation based image fusion is one of the sought after fusion techniques among the current researchers. A novel image fusion algorithm based on focused vector detection is proposed in this thesis. Firstly, the initial fused vector is acquired by combining common and innovative sparse components of multi-dosage ensemble using Joint Sparse PCA fusion method utilizing an overcomplete dictionary trained using high dose images of the same region of interest from different patients. And then, the strongly focused vector is obtained by determining the pixels of low dose and medium dose vectors which have high similarity with the pixels of the initial fused vector using certain quantitative metrics. Final fused image is obtained by denoising and simultaneously integrating the strongly focused vector, initial fused vector and source image vectors in joint sparse domain thereby preserving the edges and other critical information needed for diagnosis. This thesis demonstrates the effectiveness of the proposed algorithms when experimented on different images and the qualitative and quantitative results are compared with some of the widely used image fusion methods.

Abstract. Roadside noise barriers (RNBs) are important urban infrastructures to ensure that cities remain liveable. However, the absence of accurate and large-scale geospatial data on RNBs has impeded the increasing progress of rational urban planning, sustainable cities, and healthy environments. To address this problem, this study creates a vectorized RNB dataset in China using street view imagery and a geospatial artificial intelligence framework. First, intensive sampling is performed on the road network of each city based on OpenStreetMap, which is used as the georeference for downloading 6×106 Baidu Street View (BSV) images. Furthermore, considering the prior geographic knowledge contained in street view images, convolutional neural networks incorporating image context information (IC-CNNs) based on an ensemble learning strategy are developed to detect RNBs from the BSV images. The RNB dataset presented by polylines is generated based on the identified RNB locations, with a total length of 2667.02 km in 222 cities. Last, the quality of the RNB dataset is evaluated from two perspectives, i.e., the detection accuracy and the completeness and positional accuracy. Specifically, based on a set of randomly selected samples containing 10 000 BSV images, four quantitative metrics are calculated, with an overall accuracy of 98.61 %, recall of 87.14 %, precision of 76.44 %, and F1 score of 81.44 %. A total length of 254.45 km of roads in different cities are manually surveyed using BSV images to evaluate the mileage deviation and overlap level between the generated and surveyed RNBs. The root mean squared error for the mileage deviation is 0.08 km, and the intersection over union for overlay level is 88.08 % ± 2.95 %. The evaluation results suggest that the generated RNB dataset is of high quality and can be applied as an accurate and reliable dataset for a variety of large-scale urban studies, such as estimating the regional solar photovoltaic potential, developing 3D urban models, and designing rational urban layouts. Besides that, the benchmark dataset of the labeled BSV images can also support more work on RNB detection, such as developing more advanced deep learning algorithms, fine-tuning the existing computer vision models, and analyzing geospatial scenes in BSV. The generated vectorized RNB dataset and the benchmark dataset of labeled BSV imagery are publicly available at https://doi.org/10.11888/Others.tpdc.271914 (Chen, 2021).

In this paper, we propose the use of lustering methods for automatic counting of pedestrians in video sequences. As input, we consider the output of those detection/tracking systems that overestimate the number of targets. Clustering techniques are applied to the resulting trajectories in order to reduce the bias between the number of tracks and the real number of targets. The main hypothesis is that those trajectories belonging to the same human body are more similar than trajectories belonging to different individuals. Several data representations and different distance/similarity measures are proposed and compared, under a common hierarchical clustering framework, and both quantitative and qualitative results are presented

Light field camera depth estimation is a core technology for high-precision three-dimensional reconstruction and realistic scene reproduction. We propose a depth estimation algorithm that fuses blurry features and planar geometric structure priors, aimed at overcoming the limitations of traditional methods in neighborhood selection and mismatching in weak texture regions. First, by constructing a multi-constraint adaptive neighborhood microimage set, the microimages with the lowest blur degree are selected to calculate matching costs, and sparse feature correspondence relationships are used to propagate depth information. Second, planar prior knowledge is introduced to optimize pixel matching costs in weak texture regions, and weights are dynamically adjusted and pixel matching costs are updated during the iterative propagation process within microimages based on matching window completeness. Then, potential mismatched points are eliminated using epipolar geometric relationships. Finally, experiments were conducted using public and real-world datasets for verification and analysis. Compared with famous depth estimation algorithms, such as Zeller and BLADE, the Our method demonstrates superior performance in quantitative depth estimation metrics, scene reconstruction completeness, object edge clarity, and depth scene coverage, providing richer and more accurate depth information.

Prolonged bed rest has significant negative impacts on the human body, particularly on the cardiovascular system. To overcome adverse effects and enhance functional recovery in bedridden patients, the goal is to mobilize patients as early as possible while controlling and stabilizing their cardiovascular system. In this paper, we used a robotic tilt table that allows early mobilization by modulating body inclination and automated leg movement to control the cardiovascular variables heart rate (HR) or systolic or diastolic blood pressures (sBP, dBP). The design and use of a control system is often done with a simulation model of a plant, but the time-variant and nonlinear nature of the cardiovascular system and subject-specific responses to external stimuli makes the modeling and identification challenging. Instead, we implemented an intelligent self-learning fuzzy controller that does not need any prior knowledge about the plant. The controller modulates the body inclination in order to adjust the cardiovascular parameters, with leg movement considered as a perturbing factor to the controller. The controller performance was evaluated in six healthy subjects. Measured mean values of HR, sBP, and dBP differed from desired reference values by 1.11 beats/min, 5.10 mmHg, and 2.69 mmHg, respectively. With this new control strategy, HR and dBP could be successfully controlled within medically tolerable ranges (deviations < 2.5 beats/min and < 5 mmHg from desired values, respectively). The control of sBP was less accurate; the results suggest that simultaneous control of multiple input stimuli rather than only adaptive automatic change of the tilt table angle might improve the controllability.