EWEnda Ward

Enda Ward

*This is an algorithmically generated author record based on publicly available information.

Enda Ward has a total of 40 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Optical Simulation and Image Quality for Automotive Semantic Segmentation

    by Daniel Jakab, Joel Herrera Vázquez, Mahendar Kumbham, Brian Deegan, Tim Brophy, Reenu Mohandas, Anthony Scanlan, Enda Ward, Fiachra Collins, Ciarán Eising

    Cameras are often deployed in automated driving systems for multiple perception tasks, including object recognition. In previous work, a novel optical simulation framework, SOLAS, was proposed to simulate wide-angle and fisheye cameras on automotive data for 2D object detection. The main focus of this work is to assess semantic segmentation networks in relation to camera quality, mainly used for collision avoidance, lane marking detection, and enhanced road navigation. Characterising semantic segmentation performance across the full range of camera production tolerances is essential, given that image quality may vary across units. This study further develops the novel optical simulation framework, SOLAS, where a Pythonic ray-tracing library, KrakenOS, is adapted to simulate wide-angle surround-view optical systems across a range of optical defocus, directly in automotive scenes. Optical defocus is achieved through sensor displacement. In this paper, we propose extending SOLAS with ARDÁN, a pipeline for measuring image quality in automotive scenes. Furthermore, we demonstrate that our framework, utilising a diverse set of cameras and dense grids of Point Spread Functions (PSFs), is scalable in semantic segmentation, achieving consistent results across various datasets supported by statistical metrics. The primary metric used for performance evaluation is mean Intersection over Union (mIoU). Based on our results, the Segformer network with 24.73M parameters was the most robust across wide-angle automotive datasets, namely, nuScenes and Virtual KITTI 2.0. In nuScenes, image sharpness (cycles/pixel) dropped by 20.5% from a nominal focus, with a small Segformer performance drop from 55.21% mIoU by 0.37%. In Virtual KITTI 2.0, image sharpness dropped by 44.66% from a nominal focus, with a Segformer performance drop by 37.42%. However, for fisheye, the PIDNet model with 28.76M parameters was the most robust model across the defocus range. In Woodscape, image sharpness dropped by 53.45% from a nominal focus, with a PIDNet performance drop by 8.55%. In Parallel Domain Woodscape, image sharpness dropped by 34.46% from a nominal focus, with a PIDNet performance drop by 7.28%.

  • Open Access

    Learnable Quantum Efficiency Filters for Urban Hyperspectral Segmentation

    by Imad Ali Shah, Jiarong Li, Ethan Delaney, Enda Ward, Martin Glavin, Edward Jones, Brian Deegan

    Hyperspectral sensing provides rich spectral information for scene understanding in urban driving, but its high dimensionality poses challenges for interpretation and efficient learning. We introduce Learnable Quantum Efficiency (LQE), a physics-inspired, interpretable dimensionality reduction (DR) method that parameterizes smooth high-order spectral response functions that emulate sensor quantum efficiency curves. Unlike conventional methods or unconstrained learnable layers, LQE enforces physically motivated constraints, including a single dominant peak, smooth responses, and bounded bandwidth. This formulation yields a compact spectral representation that preserves discriminative information while remaining fully differentiable and end-to-end trainable within semantic segmentation models (SSMs). We conduct systematic evaluations across three publicly available multi-class hyperspectral urban driving datasets, comparing LQE against six conventional and seven learnable baseline DR methods across six SSMs. Averaged across all SSMs and configurations, LQE achieves the highest average mIoU, improving over conventional methods by 2.45%, 0.45%, and 1.04%, and over learnable methods by 1.18%, 1.56%, and 0.81% on HyKo, HSI-Drive, and Hyperspectral City, respectively. LQE maintains strong parameter efficiency (12–36 parameters compared to 51–22K for competing learnable approaches) and competitive inference latency. Ablation studies show that low-order configurations are optimal, while the learned spectral filters converge to dataset-intrinsic wavelength patterns. These results demonstrate that physics-informed spectral learning can improve both performance and interpretability, providing a principled bridge between hyperspectral perception and data-driven multispectral sensor design for automotive vision systems.


  • Open Access

    CSNR and JMIM Based Spectral Band Selection for Reducing Metamerism in Urban Driving

    by Jiarong Li, Imad Ali Shah, Diarmaid Geever, Fiachra Collins, Enda Ward, Martin Glavin, Edward Jones, Brian Deegan

    Protecting Vulnerable Road Users (VRU) remains a key challenge in automotive perception, particularly when RGB imagery exhibits metameric ambiguity under challenging illumination. Hyperspectral Imaging (HSI) can provide material-dependent cues beyond RGB, including in the Near-Infrared (NIR), but its high dimensionality limits deployment in automotive systems. This paper proposes a compact band-selection pipeline that combines joint mutual information maximization with a patch-based contrast signal-to-noise ratio criterion to select a practical 3-band HSI subset for VRU perception. On the Hyperspectral City V2 dataset, the proposed pipeline selects a VIS-NIR triplet at 521 nm, 753 nm, and 903 nm. We evaluate VRU-Road separability under higher-illumination scenes and an objectively defined low-illumination subset using per-scene distributions, and bootstrap confidence intervals. Results are reported in both a metric-ready representation (for direct comparison with co-registered RGB) and a sensor-native representation. Downstream utility is further evaluated by semantic segmentation using U-Net, DeepLabV3+, and PSPNet over three random seeds. The selected triplet improves VRU-Road separability on key criteria, remains robust under low illumination, and yields competitive downstream performance among the evaluated HSI-derived inputs. Overall, the proposed approach reduces spectral dimensionality from 128 bands to 3 (97.7% reduction) and supports compact multispectral sensing as a practical complement to RGB for robust VRU perception.


  • Open Access

    Impact of Inter-Modality Misalignment on Multispectral Object Detection

    by Muhammad Usama, Imad Ali Shah, Roshan George, Enda Ward, Fiachra Collins, William O’GRady, Martin Glavin, Brian Deegan, Edward Jones

    Object Detection significantly benefits from combining complementary spectral modalities; however, its performance suffers greatly from misregistration between sensor streams due to spatial misalignments, viewpoint differences between sensors, and inconsistent annotations, a common feature of multi-modal datasets used in object detection system development. In this work, we address the critical challenge of modality misregistration. We propose a comprehensive alignment framework based on object-aware local homography estimation, anchored on Long-Wave Infrared Radiation (LWIR) annotations, to generate a spatially aligned training dataset for improved model development. For our experiments, we utilise the Multispectral Object Detection (MOD) dataset, which includes RGB, Short-Wave Infrared Radiation (SWIR), Mid-Wave Infrared Radiation (MWIR), and Long-Wave Infrared Radiation (LWIR) imagery. The proposed alignment pipeline ensures a consistent field of view, resolution normalization, and accurate cross-modality annotation projection. To quantify the impact of alignment, we benchmark four state-of-the-art fusion architectures (DEYOLO, TarDAL, CMTFusion, and U2Fusion) on both the original and the aligned versions of the MOD dataset. Experimental results demonstrate significant improvements of 10 to 30 percentage points in mAP@0.5 after training on the aligned MOD data, with enhanced localization accuracy and reduced false detections. We further verified the generalizability of the proposed alignment framework on a different dataset (FLIR ADAS), observing 3.54 percentage points improvement in mAP@0.5 after training on the aligned RGB-LWIR image pairs. U2Fusion emerged as the architecture most tolerant to misregistration in initial experiments, and this architecture is used to conduct a further sensitivity analysis to evaluate model robustness against real-world misalignments, in a controlled way. The results of this sensitivity analysis could be useful for designers of practical systems, where small misalignment errors between image modalities commonly occur.


  • Open Access

    Information Capacity as a Predictor of Perception Performance

    by Diarmaid Geever, Tim Brophy, Dara Molloy, Enda Ward, Roshan George, Norman Koren, Martin Glavin, Edward Jones, Brian Deegan

    The design of automated driving systems is of growing industry and societal interest. Perception is a critical technology for these systems, which allows a vehicle to discern the surrounding environment. Perception systems in automated vehicles frequently use machine vision algorithms; however, the performance of a machine vision algorithm critically depends on the quality of the data provided. Quantifying the ‘quality’ of image data is therefore potentially a useful tool in understanding and predicting the performance of a machine vision system. This study uses the Shannon Information Capacity, a metric based on information theory, to evaluate the impact of image quality on a perception algorithm. In this preliminary study, a set of synthetic objects are arranged to create a novel simulated test chart. The chart contains standard machine vision objects of interest (people and cars) as well as a slanted edge, which is used to calculate image quality metrics. The chart is degraded using varying levels of contrast and blur to simulate different real-world operating conditions. Object detection performance is then evaluated using a range of deep learning-based detection algorithms, with different architectures. The results indicate that Shannon Information Capacity has the potential to predict machine vision performance across multiple model architectures and object types. For example, the results for all the models show that accuracy remains relatively constant above an SIC value of 0.25 b/p. Results indicate that for YOLOv10m SIC has mutual information value with detection accuracy of 1.66 bits while MTF50 has a score of 0.4945 bits. This study is the first to show the correlation between SIC and machine vision performance. While other metrics have been previously shown to have some correlation with machine vision, the correlation shown by SIC is much stronger. The findings presented may be of use to designers of autonomous driving systems and automotive camera manufacturers.


  • Open Access

    Impact of Contrast Threshold on Event-Based Automotive Perception Performance

    by Ethan Delaney, Roshan George, Tim Brophy, Imad Ai Shah, Enda Ward, Edward Jones, Brian Deegan, Martin Glavin

  • Open Access

    The Impact of Lossy Compression on Raw Bayer Images in the Automotive Environment

    by Jordan Cahill, Diarmaid Geever, Lauren Mcmahon, Dara Molloy, Enda Ward, Patrick Denny, Martin Glavin, Brian Deegan, Edward Jones

  • Open Access

    TFE-Fusion: Tri-Modality Feature Enhanced Fusion for Robust Object Detection

    by Muhammad Usama, Imad Ali Shah, Roshan George, Enda Ward, Fiachra Collins, William O'GRady, Martin Glavin, Brian Deegan, Edward Jones

  • Open Access

    G-MIND: Galway Multimodal Infrastructure Node Dataset for Intelligent Transportation Systems

    by Dara Molloy, Roshan George, Tim Brophy, Brian Deegan, Darragh Mullins, Enda Ward, Jonathan Horgan, Ciaran Eising, Patrick Denny, Edward Jones, et al.

    Autonomous and semi-autonomous vehicles require accurate perception of their surrounding environment to ensure safe operation, yet onboard sensors frequently encounter occlusion challenges that result in incomplete dynamic environmental maps. Infrastructure-to-vehicle cooperative perception addresses this by deploying infrastructure nodes that monitor scenes and share reliable environmental maps with nearby vehicles via technologies like C-V2X. However, existing infrastructure perspective datasets lack diverse multi-modal data and aerial footage, which are crucial to determine effectively the necessary sensors for safety-critical infrastructure node applications. This paper introduces G-MIND, a multimodal infrastructure node dataset supporting research into sensor suitability for infrastructure-assisted safety-critical applications. G-MIND is the first dataset to incorporate this comprehensive range of sensing modalities for infrastructure-based perception: RGB, FIR, and neuromorphic cameras, LiDARs, RADAR, and aerial drone footage. With 91,500 annotated frames, G-MIND offers a larger scale than existing infrastructure perception datasets such as Ko-PER (10k frames), CoopScenes (40K frames), and DAIR-V2X (71k frames), enabling more comprehensive training and evaluation. The dataset captures day and night scenarios featuring cars, pedestrians, and cyclists across diverse traffic scenarios. Beyond standard perception benchmarking, G-MIND includes specialized collections designed to test perception system boundaries: maximum detection distance scenarios, far and occluded object scenarios, and pedestrian action prediction scenarios that challenge current algorithms. Additionally, this paper analyzes what constitutes effective ITS infrastructure node sensors from a practical perspective, comparing modalities against technical criteria (field of view, spatial resolution, low light performance, adverse weather resilience) and pragmatic criteria (cost, durability). The dataset is available for non-commercial research at https://ieee-dataport.org/documents/galway-multimodal-infrastructure-node-dataset , with the SDK at https://github.com/daramolloy/GMIND-sdk.