Publication type: Conferences

Authors:Gonzalez Cid, Yolanda, Guerrero Tomé, Carlos, Lobo, Joao, Tarbal Roquero, Arian, Picking, Richard
Castell Ariño, NúriaMés informacióMés informació, Abdelaziz, Salih, Voegel, Bernd, Tous, Xisco, Pattichis, Constantinos, Rigaud, Bernard

authors: N. Castell1, J. Lobo2, E. Insa Calderon3, R. Picking4, Y. González5, S. Abdelaziz6, B. Voegel7, X. Tous8, C. Pattichis9, B. Rigaud10

The advent of wearable recorders poses new challenges to electrocardiogram (ECG) analysis, such as robust feature extraction in front of long-term recordings with intervals of extreme noise. This paper proposes a robust approach to improve the estimates of one particular feature, the R-R interval (RRI), extracted by an arbitrary QRS detector operating in these scenarios. The proposal performs three steps. First, a voting schema is used to detect noisy intervals. Second, a rough estimate of the RRI evolution with time is obtained. Finally, this estimate is used to guide the Kalman filter in charge of refining the RRI estimates. Two groups of experiments have been performed. The first relies on 1674 real ECG corrupted with controlled amounts of noise. The second one tests our proposal using the MIT-BIH Noise Stress Test Database. Results show that our approach is barely influenced by the initial error, leading to a large improvement in front of highly corrupted electrocardiograms at the cost of reducing the quality of the RRI estimates in absence of significant noise. Accordingly, the presented approach is suitable to process data obtained from portable ECG devices in which localized intervals of severe noise are present.

This paper presents a novel Deep Neural Network aimed at fast and robust visual loop detection targeted to underwater images. In order to help the proposed network to learn the features that define loop closings, a global image descriptor built upon clusters of local SIFT descriptors is proposed. Also, a method allowing unsupervised training is presented, eliminating the need for a hand-labelled ground truth. Once trained, the Neural Network builds two descriptors of an image that can be easily compared to other image descriptors to ascertain if they close a loop or not. The experimental results, performed using real data gathered in coastal areas of Mallorca (Spain), show the validity of our proposal and favorably compares it to previously existing methods.

{This paper constitutes a first step towards the use of Deep Neural Networks to fast and robustly detect underwater visual loops. The proposed architecture is based on an autoencoder, replacing the decoder part by a set of fully connected layers. Thanks to that it is possible to guide the training process by means of a global image descriptor built upon clusters of local SIFT features.
After training, the NN builds two different descriptors of the input image. Both descriptors can be compared among different images to decide if they are likely to close a loop. The experiments, performed in coastal areas of Mallorca (Spain), evaluate both descriptors, show the ability of the presented approach to detect loop candidates and favourably compare our proposal to a previously existing method.

This study constitutes a first step towards a wearable epileptic seizure prediction device. We exploit the existing correlation between epileptic pre-ictal states and heart rate variability features, since they can be measured by portable electrocardiogram recorders. By explicitly dealing with the intervals of extreme noise that may corrupt the electrocardiogram data during the seizures, our proposal is able to robustly train and use a Support Vector Machine to detect pre-ictal states. The experimental results show particularly good results in terms of positive and negative prediction. They also show the importance of a specific training for each patient.

This paper presents a robust approach to estimate the relative motion between couples of range scans called CSoG. The algorithm first searches prominent structural features in one of the scans by means of a clustering algorithm. Thus, no assumptions about the environment are made. Afterwards, it projects the other scan into the detected feature set and uses a score function to evaluate the projection. By optimizing the score function the motion between the two scans is obtained.

Our approach is compared to two well known scan matchers using real data from three different sensors: a terrestrial sonar, a terrestrial laser and an underwater sonar. Results show a significant improvement of CSoG with respect to the other algorithms in the case of medium and large motions between the scans. Accordingly, CSoG is a good choice to perform dead reckoning from range data and to close large loops in SLAM.