27 April 2016 in IEEE Journal of Selected Topics in Signal Processing
With increasing importance given to tele-rehabilitation, there is a growing need for accurate, low-cost, and portable motion capture systems that do not require specialist assessment venues. This paper proposes a novel framework for motion capture using only a single depth camera, which is portable and cost effective compared to most industry-standard optical systems, without compromising on accuracy. Novel signal processing and computer vision algorithms are proposed to determine motion patterns of interest from infrared and depth data. In order to demonstrate the proposed framework's suitability for rehabilitation, we developed a gait analysis application that depends on the underlying motion capture sub-system. Each subject's individual kinematics parameters, which are unique to that subject, are calculated and stored for monitoring individual progress of the clinical therapy. Experiments were conducted on 14 different subjects, 5 healthy, and 9 stroke survivors. The results show very close agreement of the resulting relevant joint angles with a benchmarking 12-camera based VICON system, a mean error of at most 1.75 % in detecting gait events w.r.t. the hand-labeled ground-truth, and significant performance improvements in terms of accuracy and efficiency compared to a previous Kinect-based system.
19 September 2019 in IEEE Transactions on Multimedia
Recent advances in image acquisition and analysis have resulted in disruptive innovation in physical rehabilitation systems facilitating cost-effective, portable, video-based gait assessment. While these inexpensive motion capture systems, suitable for home rehabilitation, do not generally provide accurate kinematics measurements on their own, image processing algorithms ensure gait analysis that is accurate enough for rehabilitation programs. This paper proposes high-accuracy classification of gait phases and muscle actions, using readings from low-cost motion capture systems. First, 12 gait parameters, drawn from the medical literature, are defined to characterize gait patterns. These proposed parameters are then used as input to our proposed multi-channel time-series classification and gait phase reconstruction methods. Proposed methods fully utilize temporal information of gait parameters, thus improving the final classification accuracy. The validation, conducted using 126 experiments, with 6 healthy volunteers and 9 stroke survivors with manually-labelled gait phases, achieves state-of-art classification accuracy of gait phase with lower computational complexity compared to previous solutions. 1
27 November 2020 in IEEE Transactions on Signal and Information Processing over Networks
Convolutional neural network (CNN)-based feature learning has become the state-of-the-art for many applications since, given sufficient training data, CNN can significantly outperform traditional methods for various classification tasks. However, feature learning is more challenging if training labels are noisy as CNN tends to overfit to the noisy training labels, resulting in sub-par classification performance. In this paper, we propose a robust binary classifier by learning CNN-based deep metric functions, to construct a graph, used to clean the noisy labels via graph Laplacian regularization (GLR). The denoised labels are then used in two proposed loss correction functions to regularize the deep metric functions. As a result, the node-to-node correlations in the graph are better reflected, leading to improved predictive performance. The experiments on three datasets, varying in number and type of features and under different levels of noise, demonstrate that given a noisy training dataset for the semi-supervised classification task, our proposed networks outperform several state-of-the-art classifiers, including label-noise robust support vector machine, CNNs with three different robust loss functions, model-based GLR, and dynamic graph CNN classifiers.
23 May 2023 in IEEE Geoscience and Remote Sensing Letters
Automatic seismic signal classification methods are extensively investigated to reduce or replace manual interpretation, with great potential in previous research. Discriminative seismic wave propagation physical characteristics, such as velocities and accelerations, are rarely considered for classification. A multitask learning scheme is proposed that utilises the seismic wave equation and three-dimensional (3D) P-wave velocity Vp model for signal representation learning. The classifier uses the obtained latent feature maps on a convolutional neural network architecture for classification of rockfall, slide quake, earthquake, and natural/anthropogenic noise events, recorded at an ongoing landslide. Our experimental results show that our approach outperforms state-of-the-art methods.
25 August 2023 in IEEE Transactions on Signal Processing
Sampling is a fundamental problem in graph signal processing that selects a node subset to collect samples, so that data in the remaining nodes can be well recovered. Existing eigen-decomposition-free (ED-free) graph sampling schemes are not designed to minimize mean square error (MMSE) of reconstructed bandlimited graph signals, often resulting in sub-par MSE performance. In this paper, we propose a lightweight ED-free algorithm to minimize an approximate MSE objective using per-node impulse responses of an ideal low-pass graph filter. Specifically, we first derive a proxy of the MMSE greedy sampling objective without matrix inverse. We then define node-dependent impulse responses of an ideal low-pass graph filter—analogous to the sinc function in traditional signal processing—which can be approximated very fast without ED. We use these response vectors to reformulate our derived sampling objective, then show that the optimal sampling set has supported low-pass impulse responses that are as orthogonal as possible—defaulting to uniform sampling for 1D regular kernels when the graph Fourier basis is DFT. For optimization, we propose a fast sampling algorithm to evaluate candidates via vector-vector multiplications by reusing previous greedy results. For faster sampling, we relax the MMSE-based greedy objective to a bounded approximation, so that candidate nodes can be easily appraised using simple scalar multiplications. Extensive experiments show that our sampling method achieved state-of-the-art sampling speed and had the best MSE performance among deterministic ED-free sampling methods in various scenarios.