No description available.

Modeling the motion that occurs between frames of a video sequence is a key component of video coding applications. Typically it is not possible to represent the motion between frames by a single model and therefore a quad-tree structure is employed where smaller, variable size regions or blocks are allowed to take on separate motion models. Quad-tree structures however suffer from two fundamental forms of redundancy. First, quad-trees exhibit structural redundancy due to their inability to exploit the dependence between neighboring leaf nodes with different parents. The second form of redundancy is due to the quad-tree structure itself being limited to capture only horizontal and vertical edge discontinuities at dyadically related locations; this means that general discontinuities in the motion field, such as those caused by boundaries of moving objects, become difficult and expensive to model. In our work, we address the issue of structural redundancy by introducing leaf merging. We describe how the intuitively appealing leaf merging step can be incorporated into quad-tree motion representations for a range motion modeling contexts. In particular, the impact of rate-distortion (R-D) optimized merging for two motion coding schemes, these being spatially predictive coding, as used by H.264, and hierarchical coding, are considered. Our experimental results demonstrate that the merging step can provide significant gains in R-D performance for both the hierarchical and spatial prediction schemes. Hierarchical coding has the advantage that it offers scalable access to the motion information; however due to the redundancy it introduces hierarchical coding has not been traditionally pursued. Our work shows that much of this redundancy can be mitigated with the introduction of merging. To enable scalable decoding, we employ a merging scheme which ensures that the dependencies introduced via merging can be hierarchically decoded. Theoretical investigations confirm the inherent advantages of leaf merging for quad-tree motion models. To enable quad-tree structures to better model motion discontinuity boundaries, we introduce geometry information to the quad-tree representation. We choose to model motion and geometry using separate quad-tree structures; thereby enabling each attribute to be refined separately. We extend the leaf merging paradigm to incorporate the dual tree structure allowing regions to be formed that have both motion and geometry attributes, subject to rate-distortion optimization considerations. We employ hierarchical coding for the motion and geometry information and ensure that the merging process retains the property of resolution scalability. Experimental results show that the R-D performance of the merged dual tree representation, is significantly better than conventional motion modeling schemes. Theoretical investigations show that if both motion and boundary geometry can be perfectly modeled, then the merged dual tree representation is able to achieve optimal R-D performance. We explore resolution scalability of merged quad-tree representations. We consider a modified Lagrangian cost function that takes into account the possibility of scalable decoding. Experimental results reveal that the new cost objective can considerably improve scalability performance without significant loss in overall efficiency and with competitive performance at all resolutions.

Estimating local site effects and modifying the uncertainty in ground motion predictions are two indispensable parts of partially non-ergodic site-specific PSHA. Local site effects can be estimated using site response simulations or recorded ground motions at the site. When such predictions are available, the aleatory variability of ground motions used in PSHA can be changed to the single station sigma value. However, in these cases, the epistemic uncertainty in predicting site effects must be incorporated into the hazard analyses. This research focuses on the challenges specific to conducting partially non-ergodic site-specific PSHA using recorded ground motions or site response analysis. The main challenge in estimating local site effects using recorded data is whether ground motions collected in a relatively short time can be used to estimate site effects for long return period events. We first develop a database for recorded ground motions at the KiK-net array to investigate this question and use this database to develop a predictive model for the Fourier Amplitude Spectra of ground motions. The ground motion model (GMM) residuals are used to investigate the stability of site terms across different tectonic regimes. We observe that empirical site terms are stable across different tectonic regimes. This observation allows the use of ground motions from any tectonic regime (whether they belong to the tectonic regime that controls the hazard or not) to estimate local site effects. Moreover, in Fourier amplitude, site effects are not dependent on event magnitude and source to site distance; therefore, estimates of site effects from low magnitude events can be easily extrapolated to larger events. The Fourier amplitude GMM developed in this study adds to the library of Fourier amplitude models to be used in future partially non-ergodic site-specific PSHAs. In practice, one of the most common tools for simulating wave propagation is 1-D site response analysis. Two central assumptions in 1-D site response analysis are that the soil profile is comprised of horizontal soil layers of infinite extent and that the vertically propagating SH-waves control the horizontal component of ground motion. SH-waves tend to propagate vertically near the surface because as earthquake waves hit softer layers traveling from the source to the site, they refract until the path becomes steeply inclined. The validity of both assumptions in 1-D site response depends on the geological setting at the site and the geology between the earthquake source and the site, raising the question of which sites are suitable for 1-D site response analysis and what the model error in 1-D site response analysis is. We use the GMM developed for FAS to estimate observed and empirical site terms. The empirical site effects are then compared with the theoretical site effects to determine whether sites are amenable to 1-D site response analyses, and to quantify the model error in the analyses. / Doctor of Philosophy / It is impossible to predict future earthquake-induced ground motions due to randomness in the process and a lack of knowledge. In fact, there are significant uncertainties not only in predicting the location, time, and magnitude of a future earthquake but also in predicting the intensity of ground motion induced by a given future earthquake. Therefore, assessing the safety of the human environment against earthquake hazards requires a method that considers all sources of uncertainties. To this end, Earthquake Engineers have developed Probabilistic Seismic Hazard Analysis(PSHA) framework. Structural engineers use the results of PSHA to design a new structure or assess the safety of an existing building. The accuracy of PSHA estimations leads to designs that are both safe and cost-efficient. The distribution of possible ground motions induced by a given earthquake scenario significantly controls the result of PSHA. This distribution should consider the effect of source, source to site path, and local site effects. This research focuses on improving PSHA results by estimating local site effects using recorded ground motions or simulating wave propagation in the site. In estimating local site effects using recorded data, the local site effect observed in ground motions collected in a relatively short time window is used to estimate hazards from all scenarios. However, the collected ground motions usually belong to frequent low magnitude events that are different from large magnitude events that control the hazard. This difference requires either using a measure of local site effect that is independent of the magnitude and distance of the earthquake or considering the effect of magnitude and distance on the local site effect estimate. Moreover, since frequent events sample different sources and paths than large events, we need to make sure the local site effect is consistent across different sources and paths. This research develops Ground Motion Models(GMMs) for Fourier amplitude, a linear function of ground motion times series, using Japanese ground motions. The ratio of Fourier amplitude at the surface over bedrock is a measure of local site effect that is not dependant on magnitude and distance. The model is then used to see if the trade-off between source and site effect and path and site effect is significant or not. In practice, one of the most common tools for simulating wave propagation is 1-D site response analysis. Two central assumptions in 1-D site response analysis are that the soil profile comprises horizontal soil layers of infinite extent and that the vertically propagating horizontal shear waves (SH-waves) control the horizontal component of ground motion. SH-waves tend to propagate vertically near the surface because as earthquake waves hit softer layers traveling from the source to the site, they refract until the path becomes vertically inclined. The validity of both assumptions in 1-D site response depends on the geological setting at the site and the geology between the earthquake source and the site, raising the question of which sites are suitable for 1-D site response analysis and what the model error in 1-D site response analysis is. We use the GMM developed for FAS to estimate empirical local site effects. The empirical site effects are then compared with the theoretical site effects to determine whether sites are amenable to 1-D site response analyses and quantify the model error in the analyses.

Video compression aims to reduce video file size without impacting visual quality. Existing algorithms mostly use transform coders to convert information from the spatial to frequency domain, and attenuate or remove high frequency components from the sequence. This enables the omission of a large proportion of high frequency information with no discernible visual impact. Sprite-based compression encodes large portions of a scene as a single object in the video sequence, recreating the object in subsequent frames by warping or morphing the sprite to mimic changes in subsequent frames. This thesis sought to improve several aspects of existing sprite based compression approaches, employing a temporal motion model using a low order polynomial to represent the motion of an object across multiple frames in a single model rather than a series of models. The main outcome is the demonstration that motion models used by sprite based video compression can be extended to a full three dimensional model, reducing the overall size of the model, and improving the quality of the sequence at low bit rates. A second outcome is the demonstration that super-resolution processing is not necessary if lanczos spatial interpolation is used instead of bilinear or bi-cubic interpolation, resulting in a savings in computational time and resources. A third outcome is the introduction of a new blending model used to generate image mosaics that improves the quality of the synthesised sequence when zoom is present in the sequence for a given bit-rate. A final outcome is demonstrating that performing superresolution processing and sub-sampling back to the original resolution prior to compression provides benefits in some circumstances.

<p>The goal of this work has been to improve the accuracy of a pre-existing algorithm for vehicle pose estimation, which uses intrinsic measurements of vehicle motion and measurements derived from far infrared images.</p><p>Estimating the pose of a vehicle, based on images from an on-board camera and intrinsic measurements of vehicle motion, is a problem of simultanoeus localization and mapping (SLAM), and it can be solved using the extended Kalman filter (EKF). The EKF is a causal filter, so if the pose estimation problem is to be solved off-line acausal methods are expected to increase estimation accuracy significantly. In this work the EKF has been compared with an acausal method for solving the SLAM problem called smoothing and mapping (SAM) which is an optimization based method that minimizes process and measurement noise.</p><p>Analyses of how improvements in the vehicle motion model, using a number of different model extensions, affects accuracy of pose estimates have also been performed.</p>

The goal of this work has been to improve the accuracy of a pre-existing algorithm for vehicle pose estimation, which uses intrinsic measurements of vehicle motion and measurements derived from far infrared images. Estimating the pose of a vehicle, based on images from an on-board camera and intrinsic measurements of vehicle motion, is a problem of simultanoeus localization and mapping (SLAM), and it can be solved using the extended Kalman filter (EKF). The EKF is a causal filter, so if the pose estimation problem is to be solved off-line acausal methods are expected to increase estimation accuracy significantly. In this work the EKF has been compared with an acausal method for solving the SLAM problem called smoothing and mapping (SAM) which is an optimization based method that minimizes process and measurement noise. Analyses of how improvements in the vehicle motion model, using a number of different model extensions, affects accuracy of pose estimates have also been performed.

With low bit rate scenarios, a hybrid video coder (e.g. AVC/H.264) tends to allocate greater portion of bits for motion vectors, while saving bits on residual errors. According to this fact, a coding scheme with non-normative global motion models in combination with conventional local motion vectors is proposed, which describes the motion of a frame by the affine motion parameter sets drawn by motion segmentation of the luminance channel. The motion segmentation task is capable of adapting the number of motion objects to the video contents. 6-D affine model sets are driven by linear regression from the scalable block-based motion fields estimated by the existent MPEG encoder. In cases that the number of motion objects exceeds a certain threshold, the global affine models are disabled. Otherwise the 4 scaling factors of the affine models are compressed by a vector quantizer, designed with a unique cache memory for efficient searching and coding. The affine motion information is written in the slice header as a syntax. The global motion information is used for compensating those macroblocks whose Lagrange cost is minimized by the AFFINE mode. The rate-distortion cost is computed by a modified Lagrange equation, which takes into consideration the perceptual discrimination of human vision in different areas. Besides increasing the coding efficiency, the global affine model manifests the following two features that refine the compressed video quality. i) When the number of slices per frame is more than 1, the global affine motion model can enhance the error-resilience of the video stream, because the affine motion parameters are duplicated in the headers of different slices over the same frame. ii) The global motion model predicts a frame by warping the whole reference frame and this helps to decrease blocking artifacts in the compensation frame.

The GPS tracking in sprint orienteering is often a poor supplement to the viewer experience during events taking place in urban areas because of multipath effects. Since the GPS tracking of runners is an important means to making the sport more spectator friendly, it is of interest to make it more accurate. In this thesis project, the information provided by the map of a competition is fused with the GPS tracker position measurements and punch time data in a particle filter to create estimates of the runner trajectories. The map is used to create constraints and to predict motion of runners, as well as to create a model of the GPS reliability depending on map position. A simple observation model is implemented, using the map to decide if a GPS measurement is reliable or not depending on the distance to the closest building. A rather complex motion model is developed to predict the runner motion within the constraints given by the map. The results show that given certain conditions the improvements are vast compared to the traditional GPS tracking. The estimates are bound to possible routes, and they are often very good given that alternative route choices are easily separable. It is however principally difficult to generally improve the tracking using this method. Better measurements or observation models are needed in order to receive a fully satisfying tracking.

No description available.

Augmented reality (AR) systems that use head mounted displays to overlay synthetic imagery on the user's view of the real world require accurate viewpoint tracking for quality applications. However, achieving accurate registration is one of the most significant unsolved problems within AR systems, particularly during dynamic motions in unprepared environments. As a result, registration error is a major issue hindering the more widespread growth of AR applications. The main objective for this thesis was to improve dynamic orientation tracking of the head using low-cost inertial sensors. The approach taken within this thesis was to extend the excellent static orientation sensing abilities of accelerometers to a dynamic case by utilising a model of head motion. Head motion is modelled by an inverted pendulum, initially for one degree of rotational freedom, but later this is extended to a more general two dimensional case by including a translational freedom of the centre of rotation. However, the inverted pendulum model consists of an unstable coupled set of differential equations which cannot be solved by conventional solution approaches. A unique method is developed which consists of a highly accurate approximated analytical solution to the full non linear tangential ODE. The major advantage of the analytical solution is that it allows a separation of the unstable transient part of the solution from the stable solution. The analytical solution is written directly in terms of the unknown initial conditions. Optimal initial conditions are found that remove the unstable transient part completely by utilising the independent radial ODE. Thus, leaving the required orientation. The methods are validated experimentally with data collected using accelerometers and a physical inverted pendulum apparatus. A range of tests were performed demonstrating the stability of the methods and solution over time and the robust performance to increasing signal frequency, over the range expected for head motion. The key advantage of this accelerometer model-based method is that the orientation remains registered to the gravitational vector, providing a drift free solution that outperforms existing, state of the art, gyroscope based methods. This proof of concept, uses low-cost accelerometer sensors to show significant potential to improve head tracking in dynamic AR environments, such as outdoors.