19 research outputs found
Extreme Precipitation Nowcasting using Deep Generative Models
No full textExtreme precipitation usually leads to substantial impacts. Floods in the Netherlands, Belgium and Germany in the summer of 2021 have caused loss of lives, destruction of infrastructures, and long-term effect on economics. To avoid such disasters, it is important to develop a reliable and accurate method to predict heavy rain.Signal Processing SystemsWater Resource
Epileptic Seizure Detection using a Tensor-Network Kalman Filter for LS-SVMs
No full textEpilepsy is one of the most common neurological conditions, affecting nearly 1% of the global population. It is defined by the seemingly random occurrence of spontaneous seizures. Anti-epileptic drugs provide adequate treatment for about 70% of patients. The remaining 30%, on the other hand, continue to have seizures, which has a significant impact on their quality of life as they are constantly unsure when these seizures will occur. Reliable seizure detection methods would thus have a significant impact on the lives of these patients. Despite ongoing research efforts involving academia and industry in large international collaborations, epileptic seizure detection and especially prediction is still an unsolved problem. The key to the solution could lie within ultralong-term, reallife datasets that are currently being generated using wearable sensors. However, due to the size of these datasets, conventional learning techniques such as least-square support vector machines (LS-SVMs) can become intractable. Therefore, this work proposes the use of a recently developed tensor network Kalman filtering approach for LS-SVMs (TNKFLSSVM) to detect epileptic seizures [1]. In the TNKF-LSSVM algorithm, the dual problem of the LS-SVM is solved using a recursive Bayesian filtering approach. This way the least-square problem can be solved row-by-row using a Kalman filter, thereby avoiding explicit matrix inversions, while also being able to provide confidence bounds on the estimates. By making use of the tensor-train format [2] to represent the matrices and vectors in the Kalman equations, it is even possible to avoid the construction of the (N + 1) × (N + 1) covariance matrix1. To be able to apply the TNKF-LSSVM algorithm for seizure detection there are still some issues that need to be tackled. One such problem is that the TNKF-LSSVM only performs well when the dataset is properly balanced, which is generally not the case for seizure datasets. Furthermore, for the TNKF-LSSVM to work efficiently for large scale problems the modes of the tensortrains representing the matrices and vectors should be as small as possible, thus it must hold that N + 1 = Q i ni, such that ni is ‘small’ for all i. To overcome both of these challenges we propose using the SMOTE method to oversample the seizure class, such that a balanced training set can be generated that has good factorization properties. Some preliminary results using a small subset of data from a public EEG dataset [3] show that taking the above considerations into account, the TNKF-LSSVM method can have performance that is competitive with a regular LS-SVM. Where the TNKFLSSVM method has the benefit of scaling log-linearly with the size of the dataset (in terms of memory usage) and can provide an uncertainty estimate of the detection. Future work will need 1N is the number of data points in the training set and 1 is added for the bias. to show whether this scaling up works as expected for the entire dataset.Signal Processing System
Aircraft Trajectory Prediction using ADS-B Data
No full textAutomatic Dependent Surveillance - Broadcast (ADS-B) is a surveillance technology that is used extensively in Air Traffic Control (ATC) applications. Aircraft equipped with ADS-B transponders actively broadcast navigation information such as position, altitude, and velocity, and thus ATC is able to track aircraft continuously, even in regions not covered by traditional radars. However, raw ADS-B messages are typically contaminated with noise, which is typically mitigated using model-based tracking methods to predict the trajectories. In this work, we propose and evaluate the performance of several filtering strategies for trajectory prediction on an existing open source TrajAir aircraft data set and our own data set i.e., collected by Delft university of technology (TUD). In our evaluation, we observe the standard Kalman filter cannot accurately track the aircraft trajectory, especially for sharply maneuvering targets. A fading-memory filter tracks maneuvering targets but introduces delay in estimates, and requires a trade-off between responsiveness and smoothness by target-specific parameter tuning. The Kalman filter with augmented process noise also involves similar trade-off and parameter tuning. Finally, the particle filter performs the best during target maneuvers but admits more noise during steady-state and increases computational cost. In this paper, we present various filtering techniques, and study the performance of these algorithms on the TrajAir and TUD aircraft data sets.Signal Processing SystemsControl & Simulatio
On the Integration of Acoustics and LiDAR: a Multi-Modal Approach to Acoustic Reflector Estimation
No full textLoudspeakers are usually placed in an environment unknown to the loudspeaker designers. Having knowledge on the room acoustic properties, e.g., the location of acoustic reflectors, allows to better reproduce the sound field as intended. Current state-of-the-art methods for room boundary detection using microphone measurements typically focus on a two-dimensional setting, causing a model mismatch when employed in real-life scenarios. Detection of arbitrary reflectors in three dimensions encounters practical limitations, e.g., the need for a spherical array and the increased computational complexity. Moreover, loudspeakers may not have an omnidirectional directivity pattern, as usually assumed in the literature, making the detection of acoustic reflectors in some directions more challenging.Signal Processing SystemsElectrical Engineering Educatio
Embedded AI Enabled Air-Writing for a Post-COVID World: Extended Abstract
No full textTouchscreens and buttons had became a medium for virus transmission during the COVID-19 pandemic. We have seen in our daily life that people use tissues and keys to press buttons inside elevators, on public screens, etc. In the post- COVID world, touch-free interaction with public touchscreens and buttons may become more popular. Motivated by the rise of visible light communication and sensing, we design a real-time embedded system to enable touch-free fingertip writing of the digits 0–9 with only ambient light and simple photodiodes. We propose an embedded deep learning model to learn the spatial and temporal patterns in the dynamic shadow for air-writing digits recognition. The model is devised with a lightweight convolutional architecture such that it can run on a resource-limited device. We evaluate our model using the LightDigit dataset [1] and report the results in terms of accuracy and inference time.Web Information SystemsEmbedded System
Image Search Engine by Deep Neural Networks
No full textWe typically search for images by keywords, e.g., when looking for images of apples, we would enter the word “apple” as query. However, there are limitations. For example, if users input keywords in a specific language, then they may miss results labeled in other languages. Moreover, users may have an image of the object they want to obtain more information about, e.g., a landmark, but they may not know the name of it. In such scenario, word-based search is not adequate, while imagebased search would be ideally suited. These needs drive us to develop a purely content-based image search engine, meaning that users can search images with an image as query. Motivated by this use case with numerous applications, in this paper we propose and validate an image query based search engine...Signal Processing System
Convergence of Stochastic PDMM
No full textIn this work, we analyse a stochastic version of the primaldual method of multipliers (PDMM), which is a promising algorithm in the field of distributed optimisation. So far, its convergence has been proven for synchronous implementations of the algorithm [1], [2]. Simulations have shown that PDMM also converges if it is implemented asynchronously, having the advantage that there is no need for clock synchronisation between the nodes in a distributed network. Furthermore, a broadcast implementation of asynchronous PDMM can be derived, instead of the usual unicast implementation. This broadcast implementation comes with a number of benefits...Electrical Engineering, Mathematics and Computer ScienceSignal Processing System
Temporal synchronization of radar and lidar streams
No full textIn multi-sensor systems, several sensors produce data streams, commonly, at different frequencies. If they are let running wild without synchronization, after a period of time, they are likely to be disordered, presenting as simultaneous measures that have been recorded at different times. That can be disastrous in many data fusion applications. This paper is about their temporal synchronization and ordering, so they can be coherently fused. Some sensors do not have timestamps from which order the streams, and even if they have, they may be not trustable for different reasons. First, we define mathematically the problem of multi-sensor data stream synchronization. Then, we handle the problem of estimating the actual time of sensor measurement using mean or median filters. Next, we address the issue of reconstructing incoming sensor data streams according to the estimated sensor measurement times while maintaining minimal latency and synchronization error by employing an adaptive stream buffering technique utilized in distributed multimedia systems. In order to test our methods, we have recorded an easy-to-use dataset with a radar and a lidar sensors without timestamps. We define a synchronization event that is easily identifiable by a human annotator in both sensor streams. From this dataset, a suitable filter for timestamp estimation is selected, and an analysis of the effects of the stream synchronization algorithm’s parameters on buffering latency and synchronization error is presented. Finally, the solution is efficiently implemented on a FPGASignal Processing System
Tensor-based Hemodynamic Response Estimation in Functional Ultrasound Data
No full textFunctional ultrasound (fUS) is an emerging technique that provides high sensitivity imaging of cerebral blood volume (CBV) changes. As increased metabolic demand of active tissue induces changes in CBV, these changes reflect neuronal activity in the corresponding brain area. The main advantages of this technique are that it can image the entire brain with unprecedented spatial (50-500um) and temporal resolution (10- 100ms), and that it constitutes a potentially portable solution, as opposed to functional magnetic resonance imaging (fMRI), the currently dominant modality in functional brain imaging. The high resolution as well as the plane-wave illumination lead to a large amount of raw ultrasound data per aquisition. The fundamental challenge is that fUS only provides an indirect measure of brain activity through the neurovascular coupling; this system is the link between the local neuronal activity and the resulting blood flow changes and has only partially known dynamic and nonlinear characteristics. Moreover, besides the activity of interest, fUS records a mixture of other ongoing brain activity, physiological artifacts and noise. The goal of this research is to develop tensor-based source separation techniques in order to estimate the brain’s hemodynamic response function (HRF) to stimuli and the activity of interest by learning its nonlinear coupling with the fUS signal.Signal Processing System
Distributed Detect-and-Avoid with Non-Stationary Obstacles
No full textDetect-and-avoid is a crucial challenge in the autonomous navigation of single or multiple agent systems. For safe and reliable autonomous navigation in unknown and dynamic environments, obstacles should be sensed using onboard sensors and the trajectory should be adjusted accordingly. Additional challenge is introduced in the case of multi-agent systems, where the adjusted trajectory could introduce collisions between agents, for example in satellite swarms in Low Earth Orbits (LEO). The increasing amount of occupancy of the low orbit and the presence of space debris gives high risk of damaging satellites due to collisions. With communication between nearby satellites, cooperative methods enable the avoidance of collisions with dynamic obstacles while simultaneously finding an optimal trajectory of the cooperative agents. Drone swarms equipped in industrial settings encounter the challenge of navigating through a dynamic environments in a similar way. The dynamic obstacles are now other autonomous systems as well as humans, performing tasks simultaneously.Signal Processing System
