A5061857063- Advanced Vision and Imaging
- Robotics and Sensor-Based Localization
- Optical measurement and interference techniques
- Computer Graphics and Visualization Techniques
- Image Enhancement Techniques
- Advanced Image Processing Techniques
- 3D Shape Modeling and Analysis
- Advanced Image and Video Retrieval Techniques
- Human Pose and Action Recognition
- Video Analysis and Summarization
- Image Processing Techniques and Applications
- Retinal and Macular Surgery
- Video Surveillance and Tracking Methods
- Human Motion and Animation
- Color Science and Applications
- Intraocular Surgery and Lenses
- Glaucoma and retinal disorders
- Image and Object Detection Techniques
- Image and Video Quality Assessment
- Advanced Neuroimaging Techniques and Applications
- Video Coding and Compression Technologies
- Visual perception and processing mechanisms
- Industrial Vision Systems and Defect Detection
- Advanced Optical Imaging Technologies
- Remote-Sensing Image Classification
University of Surrey
2016-2025
We present an approach to multi-person 3D pose estimation and tracking from multi-view video. Following independent 2D detection in each view, we: (1) correct errors the output of detector; (2) apply a fast greedy algorithm for associating detections between camera views; (3) use associated poses generate track skeletons. Previous methods estimating skeletons multiple people suffer long processing times or rely on appearance cues, reducing their applicability sports. Our views works by...
This paper introduces a general approach to dynamic scene reconstruction from multiple moving cameras without prior knowledge or limiting constraints on the structure, appearance, illumination. Existing techniques wide-baseline camera views primarily focus accurate in controlled environments, where are fixed and calibrated background is known. These approaches not robust for scenes captured with sparse cameras. Previous outdoor assume of static appearance structure. The primary contributions...
This paper presents an approach for reconstruction of 4D temporally coherent models complex dynamic scenes. No prior knowledge is required scene structure or camera calibration allowing from multiple moving cameras. Sparse-to-dense temporal correspondence integrated with joint multi-view segmentation and to obtain a complete representation static objects. Temporal coherence exploited overcome visual ambiguities resulting in improved Robust objects achieved by introducing geodesic star...
Current state-of-the-art image-based scene reconstruction techniques are capable of generating high-fidelity 3D models when used under controlled capture conditions. However, they often inadequate in more challenging outdoor environments with moving cameras. In this case, algorithms must be able to cope relatively large calibration and segmentation errors as well input images separated by a wide-baseline possibly captured at different resolutions. paper, we propose technique which, these...
Existing systems for 3-D reconstruction from multiple view video use controlled indoor environments with uniform illumination and backgrounds to allow accurate segmentation of dynamic foreground objects. In this paper, we present a portable system outdoor scenes that require relatively large capture volumes complex nonuniform illumination. This is motivated by the demand natural support film broadcast production. Limitations existing techniques in are identified. Outdoor scene performed...
We present a novel approach to 2D-3D registration from points or lines without correspondences. While there exist established solutions in the case where correspondences are known, many situations it is not possible reliably extract such across modalities, thus requiring use of correspondence-free algorithm. Existing methods rely on local search strategies and consequently have no guarantee finding optimal solution. In contrast, we first globally correspondences, achieved by Branch-and-Bound...
Stereoscopic imaging is becoming increasingly popular. However, to ensure the best quality of experience, there a need develop more robust and accurate objective metrics for stereoscopic content assessment. Existing image video are either extensions conventional 2-D (with added depth or disparity information) based on relatively simple perceptual models. Consequently, they tend lack accuracy robustness required This paper introduces full-reference human visual system (HVS) model...
Abstract Scene relighting and estimating illumination of a real scene for insertion virtual objects in mixed‐reality scenario are well‐studied challenges the computer vision graphics fields. Classical inverse rendering approaches aim to decompose into its orthogonal constituting elements, namely geometry, surface materials, which can later be used augmented reality or render new images under novel lighting viewpoints. Recently, application deep neural computing estimation, has shown...
A 4D parametric motion graph representation is presented for interactive animation from actor performance capture in a multiple camera studio. The based on model database of temporally aligned mesh sequence reconstructions motions. High-level movement controls such as speed and direction are achieved by blending sequences related real-time approach introduced which combines the realistic deformation previous non-linear solutions with efficient online computation. Transitions between...
A 4D parametric motion graph representation is presented for interactive animation from actor performance capture in a multiple camera studio. The based on model database of temporally aligned mesh sequence reconstructions motions. High-level movement controls such as speed and direction are achieved by blending sequences related real-time approach introduced, which combines the realistic deformation previous nonlinear solutions with efficient online computation. Transitions between...
Human pose estimation from monocular video streams is a challenging problem. Much of the work on this problem has focused developing inference algorithms and probabilistic prior models based learned measurements. Such face challenges in generalization beyond dataset. We propose an interactive model-based generative approach for estimating human 2D uncalibrated unconstrained sports TV footage without any learning motion captured or annotated data. Belief-propagation over spatio-temporal graph...
To study the behaviour of expansile intravitreal gases and air used in treating rhegmatogenous retinal detachment.A validated mathematical model gas expansion absorption human eyes was to simulate effect varying volumes pure air, SF6, C2F6 C3F8 injected into vitreous cavity. Variation axial length accounted for by using three different cavity represent hypermetropic, emmetropic myopic eyes.The time course fluorinated were tabulated, with calculated parameters including volume gas, percentage...
3DTV production of live sports events presents a challenging problem involving conflicting requirements maintaining broadcast stereo picture quality with practical problems in developing robust systems for cost effective deployment. In this paper we propose an alternative approach to using the conventional monocular cameras 3D reconstruction event and subsequent rendering. This has potential advantage over camera rigs recovering full scene depth, allowing inter-ocular distance convergence be...
We present a family of methods for 2D–3D registration spanning both deterministic and non-deterministic branch-and-bound approaches. Critically, the exhibit invariance to underlying scene primitives, enabling e.g. points lines be treated on an equivalent basis, potentially broader range problems tackled while maximising available information, all primitives being simultaneously considered. Being based approach, method furthermore enjoys intrinsic guarantees global optimality; approaches have...
This paper addresses the problem of human action matching in outdoor sports broadcast environments, by analysing 3D data from a recorded activity and retrieving most appropriate proxy motion capture library. Typically pose recognition is carried out using images single camera, however this approach sensitive to occlusions restricted fields view, both which are common environment. presents novel technique for automatic activities operates on available multi-camera Shape retrieved techniques...
Here we present a novel, information-theoretic salient line segment detector. Existing detectors typically only use the image gradient to search for potential lines. Consequently, many lines are found, particularly in repetitive scenes. In contrast, our approach detects that define regions of significant divergence between pixel intensity or colour statistics. This results novel detector naturally avoids parts scene while detecting strong, discriminative present. We furthermore as saliency...
Stereoscopic video quality assessment has become a major research topic in recent years. Existing stereoscopic metrics are predominantly based on image extended to the time domain via for example temporal pooling. These approaches do not explicitly consider motion sensitivity of Human Visual System (HVS). To address this limitation, paper introduces novel HVS model inspired by physiological findings characterising sensitive response complex cells primary visual cortex (V1 area). The proposed...
Conventional approaches to 3D scene reconstruction often treat matting and as two separate problems, with a prerequisite reconstruction. The problem such an approach is that it requires taking irreversible decisions at the first stage, which may translate into errors second stage. In this paper, we propose attempts solve both problems jointly, thereby avoiding limitation. A general Bayesian formulation for estimating opacity depth respect reference camera developed. addition, demonstrated in...
We present a method for reconstructing the geometry and appearance of indoor scenes containing dynamic human subjects using single (optionally moving) RGBD sensor. introduce framework building representation articulated scene as set piecewise rigid parts which are tracked accumulated over time moving voxel grids signed distance representation. Data association noisy depth measurements with body is achieved by online training prior shape model specific subject. A novel frame-to-frame...
Light-field video has recently been used in virtual and augmented reality applications to increase realism immersion. However, existing light-field methods are generally limited static scenes due the requirement acquire a dense scene representation. The large amount of data absence infer temporal coherence pose major challenges storage, compression editing compared conventional video. In this paper, we propose first method extract spatio-temporally coherent A novel obtain Epipolar Plane...