Deep Learning Methods for Speed Estimation of Bipedal Motion from Wearable IMU Sensors.

The estimation of the speed of human motion from wearable IMU sensors is required in applications such as pedestrian dead reckoning. In this paper, we test deep learning methods for the prediction of the motion speed from raw readings of a low-cost IMU sensor. Each subject was observed using three sensors at the shoe, shin, and thigh. We show that existing general-purpose architectures outperform classical feature-based approaches and propose a novel architecture tailored for this task. The proposed architecture is based on a semi-supervised variational auto-encoder structure with innovated decoder in the form of a dense layer with a sinusoidal activation function. The proposed architecture achieved the lowest average error on the test data. Analysis of sensor placement reveals that the best location for the sensor is the shoe. Significant accuracy gain was observed when all three sensors were available. All data acquired in this experiment and the code of the estimation methods are available for download.

Comparison of Anomaly Detectors: Context Matters.

Deep generative models are challenging the classical methods in the field of anomaly detection nowadays. Every newly published method provides evidence of outperforming its predecessors, sometimes with contradictory results. The objective of this article is twofold: to compare anomaly detection methods of various paradigms with a focus on deep generative models and identification of sources of variability that can yield different results. The methods were compared on popular tabular and image datasets. We identified that the main sources of variability are the experimental conditions: 1) the type of dataset (tabular or image) and the nature of anomalies (statistical or semantic) and 2) strategy of selection of hyperparameters, especially the number of available anomalies in the validation set. Methods perform differently in different contexts, i.e., under a different combination of experimental conditions together with computational time. This explains the variability of the previous results and highlights the importance of careful specification of the context in the publication of a new method. All our code and results are available for download.

Source term determination with elastic plume bias correction.

Estimation of a source term, i.e. release rate, of atmospheric radionuclide emissions is of key interest for nuclear emergency response and further accident analysis. The source term estimate is, however, often very inaccurate due to biases in atmospheric transport and used meteorological analysis. We propose a method for atmospheric plume bias correction which uses not only concentrations modeled at a measuring site but also the information on concentration gradient from the neighborhood of each measuring site, i.e. information already available from the atmospheric transport model. To properly regularize the model, we propose an elastic model of the plume bias correction based on regularization with the use of known topology of the measurement network. The proposed plume bias correction method can be coupled with an arbitrary source term estimation algorithm and can be instantly applied to any other atmospheric release of hazardous material. We demonstrate the method in two real cases. First, we use data from the European Tracer Experiment to validate the methodology. Second, we use data from the 106Ru occurrence over Europe in 2017 to demonstrate the methodology in a more demanding case where agreement with state-of-the-art estimates is shown with much better reconstruction of measurements.

Fast AHRS Filter for Accelerometer, Magnetometer, and Gyroscope Combination with Separated Sensor Corrections.

A new predictor-corrector filter for attitude and heading reference systems (AHRS) using data from an orthogonal sensor combination of three accelerometers, three magnetometers and three gyroscopes is proposed. The filter uses the predictor-corrector structure, with prediction based on gyroscopes and independent correction steps for acceleration and magnetic field sensors. We propose two variants of the filter: (i) one using mathematical operations of special orthogonal group SO(3), that are accurate for nonlinear operations, for highest possible accuracy, and (ii) one using linearization of nonlinear operations for fast evaluation. Both approaches are quaternion-based filter realizations without redundant steps. The filters are compared to state of the art methods in this field on data recorded using low-cost microelectromechanical systems (MEMS) sensors with ground truth measured by the VICON optical system. Both filters achieved better accuracy than conventional methods at lower computational cost. The recorded data with ground truth reference and the source codes of both filters are publicly available.

Blind Deconvolution With Model Discrepancies.

Blind deconvolution is a strongly ill-posed problem comprising of simultaneous blur and image estimation. Recent advances in prior modeling and/or inference methodology led to methods that started to perform reasonably well in real cases. However, as we show here, they tend to fail if the convolution model is violated even in a small part of the image. Methods based on variational Bayesian inference play a prominent role. In this paper, we use this inference in combination with the same prior for noise, image, and blur that belongs to the family of independent non-identical Gaussian distributions, known as the automatic relevance determination prior. We identify several important properties of this prior useful in blind deconvolution, namely, enforcing non-negativity of the blur kernel, favoring sharp images over blurred ones, and most importantly, handling non-Gaussian noise, which, as we demonstrate, is common in real scenarios. The presented method handles discrepancies in the convolution model, and thus extends applicability of blind deconvolution to real scenarios, such as photos blurred by camera motion and incorrect focus.

Inverse modelling for real-time estimation of radiological consequences in the early stage of an accidental radioactivity release.

A stepwise sequential assimilation algorithm is proposed based on an optimisation approach for recursive parameter estimation and tracking of radioactive plume propagation in the early stage of a radiation accident. Predictions of the radiological situation in each time step of the plume propagation are driven by an existing short-term meteorological forecast and the assimilation procedure manipulates the model parameters to match the observations incoming concurrently from the terrain. Mathematically, the task is a typical ill-posed inverse problem of estimating the parameters of the release. The proposed method is designated as a stepwise re-estimation of the source term release dynamics and an improvement of several input model parameters. It results in a more precise determination of the adversely affected areas in the terrain. The nonlinear least-squares regression methodology is applied for estimation of the unknowns. The fast and adequately accurate segmented Gaussian plume model (SGPM) is used in the first stage of direct (forward) modelling. The subsequent inverse procedure infers (re-estimates) the values of important model parameters from the actual observations. Accuracy and sensitivity of the proposed method for real-time forecasting of the accident propagation is studied. First, a twin experiment generating noiseless simulated "artificial" observations is studied to verify the minimisation algorithm. Second, the impact of the measurement noise on the re-estimated source release rate is examined. In addition, the presented method can be used as a proposal for more advanced statistical techniques using, e.g., importance sampling.

Bayesian blind separation and deconvolution of dynamic image sequences using sparsity priors.

A common problem of imaging 3-D objects into image plane is superposition of the projected structures. In dynamic imaging, projection overlaps of organs and tissues complicate extraction of signals specific to individual structures with different dynamics. The problem manifests itself also in dynamic tomography as tissue mixtures are present in voxels. Separation of signals specific to dynamic structures belongs to the category of blind source separation. It is an underdetermined problem with many possible solutions. Existing separation methods select the solution that best matches their additional assumptions on the source model. We propose a novel blind source separation method based on probabilistic model of dynamic image sequences assuming each source dynamics as convolution of an input function and a source specific kernel (modeling organ impulse response or retention function). These assumptions are formalized as a Bayesian model with hierarchical prior and solved by the Variational Bayes method. The proposed prior distribution assigns higher probability to sparse source images and sparse convolution kernels. We show that the results of separation are relevant to selected tasks of dynamic renal scintigraphy. Accuracy of tissue separation with simulated and clinical data provided by the proposed method outperformed accuracy of previously developed methods measured by the mean square and mean absolute errors of estimation of simulated sources and the sources separated by an expert physician. MATLAB implementation of the algorithm is available for download.