Department Publications

Manual annotation of large-scale point cloud dataset for varying tasks such as 3D object classification, segmentation and detection is often laborious owing to the irregular structure of point clouds. Self-supervised learning, which operates without any human labeling, is a promising approach to address this issue. We observe in the real world that humans are capable of mapping the visual concepts learnt from 2D images to understand the 3D world. Encouraged by this insight, we propose CrossPoint, a simple cross-modal contrastive learning approach to learn transferable 3D point cloud representations. It enables a 3D-2D correspondence of objects by maximizing agreement between point clouds and the corresponding rendered 2D image in the invariant space, while encouraging invariance to transformations in the point cloud modality. Our joint training objective combines the feature correspondences within and …

This paper presents a sound source localization method using an irregular microphone array embedded in a drone. Sound source localization is an integral function of drone audition systems which enables various applications of drones such as search and rescue missions. However, the audio recordings using the on-board microphones obscure the sound emitted by a source on the ground due to drone generated motor and propeller noise, thus leading to an extremely low signal-to-drone noise ratio (SdNR). In this paper, we propose a cross-correlation based direction of arrival (DOA) estimation technique using the time difference of arrival (TDOA) at different microphone pairs, with noise angular spectrum subtraction. Through the measured current-specific drone noise spectrum, noise suppression has been achieved from the multi-channel recordings. Experimental results show that the proposed method is capable …

In this paper, we introduce a novel road marking bench-mark dataset for road marking detection, addressing the limitations in the existing publicly available datasets such as lack of challenging scenarios, prominence given to lane markings, unavailability of an evaluation script, lack of an-notation formats and lower resolutions. Our dataset consists of 2887 total images with 4706 road marking instances belonging to 11 classes. The images have a high resolution of 1920 × 1080 and capture a wide range of traffic, lighting and weather conditions. We provide road marking an-notations in polygons, bounding boxes and pixel-level segmentation masks to facilitate a diverse range of road marking detection algorithms. The evaluation metrics and the evaluation script we provide, will further promote direct comparison of novel approaches for road marking detection with existing methods. Furthermore, we evaluate the …

Recent work done on traffic sign and traffic light detection focus on improving detection accuracy in complex scenarios, yet many fail to deliver real-time performance, specifically with limited computational resources. In this work, we propose a simple deep learning based end-to-end detection framework, which effectively tackles challenges inherent to traffic sign and traffic light detection such as small size, large number of classes and complex road scenarios. We optimize the detection models using TensorRT and integrate with Robot Operating System to deploy on an Nvidia Jetson AGX Xavier as our embedded device. The overall system achieves a high inference speed of 63 frames per second, demonstrating the capability of our system to perform in real-time. Furthermore, we introduce CeyRo, which is the first ever large-scale traffic sign and traffic light detection dataset for the Sri Lankan context. Our dataset …

Objective The Temporal Response Function (TRF) is a linear model of neural activity time-locked to continuous stimuli, including continuous speech. TRFs based on speech envelopes typically have distinct components that have provided remarkable insights into the cortical processing of speech. However, current methods may lead to less than reliable estimates of single-subject TRF components. Here, we compare two established methods, in TRF component estimation, and also propose novel algorithms that utilize prior knowledge of these components, bypassing the full TRF estimation. Methods We compared two established algorithms, ridge and boosting, and two novel algorithms based on Subspace Pursuit (SP) and Expectation Maximization (EM), which directly estimate TRF components given plausible assumptions regarding component characteristics. Single-channel, multi-channel, and source-localized …

Stroke patients with hemiparesis display decreased beta band (13–25 Hz) rolandic activity, correlating to impaired motor function. However, clinically, patients without significant weakness, with small lesions far from sensorimotor cortex, exhibit bilateral decreased motor dexterity and slowed reaction times. We investigate whether these minor stroke patients also display abnormal beta band activity. Magnetoencephalographic (MEG) data were collected from nine minor stroke patients (NIHSS < 4) without significant hemiparesis, at ~1 and ~6 months postinfarct, and eight age-similar controls. Rolandic relative beta power during matching tasks and resting state, and Beta Event Related (De)Synchronization (ERD/ERS) during button press responses were analyzed. Regardless of lesion location, patients had significantly reduced relative beta power and ERS compared to controls. Abnormalities persisted over visits, and were present in both ipsi- and contra-lesional hemispheres, consistent with bilateral impairments in motor dexterity and speed. Minor stroke patients without severe weakness display reduced rolandic beta band activity in both hemispheres, which may be linked to bilaterally impaired dexterity and processing speed, implicating global connectivity dysfunction affecting sensorimotor cortex independent of lesion location. Findings not only illustrate global network disruption after minor stroke, but suggest rolandic beta band activity may be a potential biomarker and treatment target, even for minor stroke patients with small lesions far from sensorimotor areas.

This paper proposes a single-channel speech enhancement method to reduce the noise and enhance speech at low signal-to-noise ratio (SNR) levels and non-stationary noise conditions. Specifically, we focus on modeling the noise using a Gaussian mixture model (GMM) based on a multi-stage process with a parametric Wiener filter. The proposed noise model estimates a more accurate noise power spectral density (PSD), and allows for better generalization under different noise conditions such as harmonic and babble noise environments compared to traditional Wiener filtering methods. Simulations show that the proposed approach can achieve better performance in terms of speech quality (PESQ) and intelligibility (STOI) at low SNR levels.

This paper addresses the problem of deploying unmanned aerial vehicles (UAVs) to assist data collection and transmission in wireless sensor networks (WSNs). A multi-UAV deployment algorithm is proposed for an energy-constrained network to minimize the maximum transmit power of the sensor node (SN), while satisfying data rate and transmit power constraints. The problem is divided into two subproblems, namely, UAV-SN association, and UAV positioning. First, the UAV-SN association is addressed using a customized Gale-Shapley algorithm. Then, the 2D positions of the UAVs are decided using a modified pattern search algorithm. The two subproblems are solved alternatively in an iterative manner to obtain a near-optimal solution. Compared to baseline algorithms, the proposed solution results in at least 4-fold and 4.5-fold reductions in the number of iterations required for convergence, and the maximum …

Human behavioral monitoring during sleep is essential for various medical applications. Majority of the contactless human pose estimation algorithms are based on RGB modality, causing ineffectiveness in in-bed pose estimation due to occlusions by blankets and varying illumination conditions. Long-wavelength infrared (LWIR) modality based pose estimation algorithms overcome the aforementioned challenges; however, ground truth pose generations by a human annotator under such conditions are not feasible. A feasible solution to address this issue is to transfer the knowledge learned from images with pose labels and no occlusions, and adapt it towards real world conditions (occlusions due to blankets). In this paper, we propose a novel learning strategy comprises of two-fold data augmentation to reduce the cross-domain discrepancy and knowledge distillation to learn the distribution of unlabeled images in …

The number of coefficients of multi-dimensional (M-D) finite-extent impulse response (FIR) filters increases exponentially with the number of dimensions leading to significantly high computational complexities. In this brief, we propose a minimax design method for M-D FIR filters having sparse coefficients, therefore, having low computational complexities. We consider the design of M-D FIR filters with arbitrary frequency responses and low group delays of which the coefficients are complex valued. We formulate the minimax design as a second-order cone programming problem. Design examples confirm that M-D sparse FIR filters designed using the proposed method provide more than 60% reduction in the computational complexity for a similar error in the frequency response approximation compared to M-D FIR nonsparse filters designed using previously proposed minimax methods.

Transformers combined with convolutional encoders have been recently used for hand gesture recognition (HGR) using micro-Doppler signatures. In this letter, we propose a vision-transformer-based architecture for HGR with multiantenna continuous-wave Doppler radar receivers. The proposed architecture consists of three modules: 1) a convolutional encoder–decoder, 2) an attention module with three transformer layers, and 3) a multilayer perceptron. The novel convolutional decoder helps to feed patches with larger sizes to the attention module for improved feature extraction. Experimental results obtained with a dataset corresponding to a two-antenna continuous-wave Doppler radar receiver operating at 24 GHz (published by Skaria et al.) confirm that the proposed architecture achieves an accuracy of 98.3% which substantially surpasses the state-of-the-art on the used dataset.

Few-Shot learning aims to train and optimize a model that can adapt to unseen visual classes with only a few labeled examples. The existing few-shot learning (FSL) methods, heavily rely only on visual data, thus fail to capture the semantic attributes to learn a more generalized version of the visual concept from very few examples. However, it is a known fact that human visual learning benefits immensely from inputs from multiple modalities such as vision, language, and audio. Inspired by the human learning nature of encapsulating the existing knowledge of a visual category which is in the form of language, we introduce a contrastive alignment mechanism for visual and semantic feature vectors to learn much more generalized visual concepts for few-shot learning. Our method simply adds an auxiliary contrastive learning objective which captures the contextual knowledge of a visual category from a strong textual encoder in addition to the existing training mechanism. Hence, the approach is more generalized and can be plugged into any existing FSL method. The pre-trained semantic feature extractor (learned from a large-scale text corpora) we use in our approach provides a strong contextual prior knowledge to assist FSL. The experimental results done in popular FSL datasets show that our approach is generic in nature and provides a strong boost to the existing FSL baselines.

We address the detection of material defects, which are inside a layered material structure using compressive sensing-based multiple-input and multiple-output (MIMO) wireless radar. Here, strong clutter due to the reflection of the layered structure’s surface often makes the detection of the defects challenging. Thus, sophisticated signal separation methods are required for improved defect detection. In many scenarios, the number of defects that we are interested in is limited, and the signaling response of the layered structure can be modeled as a low-rank structure. Therefore, we propose joint rank and sparsity minimization for defect detection. In particular, we propose a non-convex approach based on the iteratively reweighted nuclear and ℓ1-norm (a double-reweighted approach) to obtain a higher accuracy compared to the conventional nuclear norm and ℓ1-norm minimization. To this end, an iterative algorithm is designed to estimate the low-rank and sparse contributions. Further, we propose deep learning-based parameter tuning of the algorithm (i.e., algorithm unfolding) to improve the accuracy and the speed of convergence of the algorithm. Our numerical results show that the proposed approach outperforms the conventional approaches in terms of mean squared errors of the recovered low-rank and sparse components and the speed of convergence.

We propose a minimax design method for finite impulse response (FIR) graph filters in this brief. We first convert the graph filter design problem into a numerically stable equivalent problem using a shifted monomial basis. Then, we convert the minimax graph filter design problem into a semidefinite programming problem. Furthermore, we propose a computationally-efficient structure to implement the FIR graph filters designed using the proposed method. Design examples confirm that the proposed method can be used to design FIR graph filters having reasonably high orders, and the proposed implementation structure substantially saves additions required to process a graph signal compared to graph filters designed using a previously proposed Chebyshev polynomial approximation method.

Any clustering algorithm must synchronously learn to model the clusters and allocate data to those clusters in the absence of labels. Mixture model-based methods model clusters with pre-defined statistical distributions and allocate data to those clusters based on the cluster likelihoods. They iteratively refine those distribution parameters and member assignments following the Expectation-Maximization (EM) algorithm. However, the cluster representability of such hand-designed distributions that employ a limited amount of parameters is not adequate for most real-world clustering tasks. In this paper, we realize mixture model-based clustering with a neural network where the final layer neurons, with the aid of an additional transformation, approximate cluster distribution outputs. The network parameters pose as the parameters of those distributions. The result is an elegant, much-generalized representation of clusters …

There is a growing demand for novel nonisolated dc–dc power converters with the deployment of dc power distribution systems integrating renewable sources and energy storing elements. The load and source of these applications have different input and output voltage levels, and it is vital to avoid operating the interfacing power converters at extreme duty ratios. Therefore, engineers and researchers explore the power converter synthesizing techniques since the mid-1970s, fulfilling the above requirement besides the nonpulsating input and output current. Among them, the analytical synthesis method has gained popularity, although it gives rise to high-entropy equations. This article shows a simple yet powerful topology synthesis method based on the low-entropy equations that reveal the connection between the energy storing elements and switches. To this end, a set of design rules have been introduced in this …

Network slicing enables the creation of multiple logical independent networks on physical networking infrastructure. Network slice deployment in a Fifth Generation (5G) mobile network can be classified as vertical and horizontal slicing. The paper compares the performance of the two slicing methods through solving two convex optimization problems, considering several smart hospital scenarios that differ from each other based on their medical speciality. The results are used to draw insights on the most appropriate slicing approach for each setup.

Raw point cloud processing using capsule networks is widely adopted in classification, reconstruction, and segmentation due to its ability to preserve spatial agreement of the input data. However, most of the existing capsule based network approaches are computationally heavy and fail at representing the entire point cloud as a single capsule. We address these limitations in existing capsule network based approaches by proposing PointCaps, a novel convolutional capsule architecture with parameter sharing. Along with PointCaps, we propose a novel Euclidean distance routing algorithm and a class-independent latent representation. The latent representation captures physically interpretable geometric parameters of the point cloud, with dynamic Euclidean routing, PointCaps well-represents the spatial (point-to-part) relationships of points. PointCaps has a significantly lower number of parameters and requires a …

Primary auditory cortex is a critical stage in the human auditory pathway, a gateway between subcortical and higher-level cortical areas. Receiving the output of all subcortical processing, it sends its output on to higher-level cortex. Non-invasive physiological recordings of primary auditory cortex using electroencephalography (EEG) and magnetoencephalography (MEG), however, may not have sufficient specificity to separate responses generated in primary auditory cortex from those generated in underlying subcortical areas or neighboring cortical areas. This limitation is important for investigations of effects of top-down processing (e.g., selective-attention-based) on primary auditory cortex: higher-level areas are known to be strongly influenced by top-down processes, but subcortical areas are often assumed to perform strictly bottom-up processing. Fortunately, recent advances have made it easier to isolate the neural activity of primary auditory cortex from other areas. In this perspective, we focus on time-locked responses to stimulus features in the high gamma band (70–150 Hz) and with early cortical latency (∼40 ms), intermediate between subcortical and higher-level areas. We review recent findings from physiological studies employing either repeated simple sounds or continuous speech, obtaining either a frequency following response (FFR) or temporal response function (TRF). The potential roles of top-down processing are underscored, and comparisons with invasive intracranial EEG (iEEG) and animal model recordings are made. We argue that MEG studies employing continuous speech stimuli may offer particular benefits, in that …

Ever since the first microscope by Zacharias Janssen in the late 16th century, scientists have been inventing new types of microscopes for various tasks. Inventing a novel architecture demands years, if not decades, worth of scientific experience and creativity. In this work, we introduce Differentiable Microscopy (∂ µ), a deep learning-based design paradigm, to aid scientists design new interpretable microscope architectures. Differentiable microscopy first models a common physics-based optical system however with trainable optical elements at key locations on the optical path. Using pre-acquired data, we then train the model end-to-end for a task of interest. The learnt design proposal can then be simplified by interpreting the learnt optical elements. As a first demonstration, based on the optical 4-f system, we present an all-optical quantitative phase microscope (QPM) design that requires no computational post-reconstruction. A follow-up literature survey suggested that the learnt architecture is similar to the generalized phase concept developed two decades ago. We then incorporate the generalized phase contrast concept to simplify the learning procedure. Furthermore, we also demonstrate that similar results can be achieved by an uninterpretable learning based method, namely diffractive deep neural networks (D2NN). We outperform the existing benchmark for all-optical phase-to-intensity conversion on multiple datasets, and ours is the first demonstration of its kind on D2NNs. The proposed differentiable microscopy framework supplements the creative process of designing new optical systems and would perhaps lead to unconventional …

Indoor cycling is a concept that has emerged with technological advancements to serve multiple purposes such as fitness, virtual tourism, and leisure. Commercial solutions have been developed to serve different user groups, with emphasis on corresponding aspects of user experience. This paper proposes a novel approach to indoor cycling, combining key features including real-world 360-degree view, immersive environment, real-time interaction between the real and virtual worlds and dynamic resistance control to simulate changes in elevation. In the proposed solution, a mobile application provides an immersive cycling environment by playing 360-degree videos on a Virtual Reality headset. The environment responds to user movements such as head rotation and cycling speed changes, and in return replicates the actual resistance to the rider using automatic resistance control. These components are …

Penetration of Intelligent Transport Systems (ITS) applications is slow, particularly in developing countries, mainly due to the costs associated with the two state-of-the-art technologies, Dedicated Short-Range Communication (DSRC) and Cellular-V2x (c-V2x).As a solution, this paper presents a DSRC/Wi-Fi hybrid system and evaluates its performance experimentally. Communication within the Wi-Fi segment is via customized beacons serving a similar function as roadside alerts in DSRC, but with a configurable repetition interval. The results show that Wi-Fi beacons can be received over a 200m range with a packet delivery rate of more than 50%. Considering the effective range and the information update rate, we illustrate several applications that are feasible with the hybrid system in Vehicle-to-Vehicle (V2V) and Infrastructure-to-Vehicle (I2V) modes. With our novel hardware design, the expected functionality can …

Inspired by recent improvements in point cloud processing for autonomous navigation, we focus on using hierarchical graph neural networks for processing and feature learning over large-scale outdoor LiDAR point clouds. We observe that existing GNN based methods fail to overcome challenges of scale and irregularity of points in outdoor datasets. Addressing the need to preserve structural details while learning over a larger volume efficiently, we propose Hierarchical Point Graph Neural Network (HPGNN). It learns node features at various levels of graph coarseness to extract information. This enables to learn over a large point cloud while retaining fine details that existing pointlevel graph networks struggle to achieve. Connections between multiple levels enable a point to learn features in multiple scales, in a few iterations. We design HPGNN as a purely GNN-based approach, so that it offers modular …

We propose a novel multi-dimensional zero-phase infinite impulse response (IIR) notch filter having multiple independently-tunable notches. The proposed notch filter employs a cascade structure with M N-dimensional ( N -D) zero-phase IIR filters to realize an N -D notch filter with M notches. In the proposed filter, the frequency and the bandwidth of each notch can be independently tuned. Furthermore, the proposed filter can have odd number of notches, which may be required for complex-valued signals. Experimental results confirm that the proposed filter significantly attenuates multiple narrowband interference and provides significant improvement in the signal-to-interference ratio compared to a previously proposed three-dimensional IIR filter.

Mostly-online teaching experiences in circuits and systems (CAS) during 2020-2021 COVID-19 pandemic are presented. Three case studies are shared summarizing course details, tools and platforms, best practices and limitations across three universities representing different continents. The presented approaches attempt to address several limitations of passive online delivery of CAS courses via interactive simulations, formative assessments via interactive web content and slide-embedded polls, at-home labs with student acquired hardware and instructor-lead and student-centered activity based learning.