No description available.

Today automotive companies across the world strive to create vehicles with fully autonomous capabilities. There are many benefits of developing autonomous vehicles, such as reduced traffic congestion, increased safety and reduced pollution, etc. To be able to achieve that goal there are many challenges ahead, one of them is visual perception. Being able to estimate depth from a 2D image has been shown to be a key component for 3D recognition, reconstruction and segmentation. Being able to estimate depth in an image from a monocular camera is an ill-posed problem since there is ambiguity between the mapping from colour intensity and depth value. Depth estimation from stereo images has come far compared to monocular depth estimation and was initially what depth estimation relied on. However, being able to exploit monocular cues is necessary for scenarios when stereo depth estimation is not possible. We have presented a novel CNN network, BiNet which is inspired by ENet, to tackle depth estimation of moving objects using only a monocular camera in real-time. It performs better than ENet in the Cityscapes dataset while adding only a small overhead to the complexity. / I dag strävar bilföretag över hela världen för att skapa fordon med helt autonoma möjligheter. Det finns många fördelar med att utveckla autonoma fordon, såsom minskad trafikstockning, ökad säkerhet och minskad förorening, etc. För att kunna uppnå det målet finns det många utmaningar framåt, en av dem är visuell uppfattning. Att kunna uppskatta djupet från en 2D-bild har visat sig vara en nyckelkomponent för 3D-igenkännande, rekonstruktion och segmentering. Att kunna uppskatta djupet i en bild från en monokulär kamera är ett svårt problem eftersom det finns tvetydighet mellan kartläggningen från färgintensitet och djupvärde. Djupestimering från stereobilder har kommit långt jämfört med monokulär djupestimering och var ursprungligen den metod som man har förlitat sig på. Att kunna utnyttja monokulära bilder är dock nödvändig för scenarier när stereodjupuppskattning inte är möjligt. Vi har presenterat ett nytt nätverk, BiNet som är inspirerat av ENet, för att ta itu med djupestimering av rörliga objekt med endast en monokulär kamera i realtid. Det fungerar bättre än ENet med datasetet Cityscapes och lägger bara till en liten kostnad på komplexiteten.

The design and maintenance of Marine Oil and LNG Terminals is governed by the Marine Oil Terminal Engineering and Maintenance Standards (MOTEMS) which is part of the 2010 Title 24 California Code of Regulations, Part 2, California Building Code, Chapter 31F: Marine Oil Terminals. The purpose of this thesis is to evaluate the current recommendations for the in-ground plastic hinge length and depth for piles in section 7 of MOTEMS for all typical soil properties and pile dimensions found in Marine Oil and LNG Terminals. The pile types considered in this analysis are 24-inch octagonal prestressed concrete piles and 24-, 36-, and 48-inch steel pipe piles in varying soil conditions. Existing recommendations for plastic hinges are incomplete and inadequate. MOTEMS does not have any recommendations for plastic hinge depth, only length, nor does it have any recommendations for in-ground plastic hinge for steel piles. Recommendations for steel piles are however found in the Port of Long Beach Wharf Design Criteria (POLB), but the recommendations in both MOTEMS and POLB have shown to be inadequate for both steel and prestressed concrete piles. MOTEMS also proves to be adequate for Level 2 earthquakes but not for Level 1. The plastic hinge length for Level 1 is much longer than that for Level 2. So the MOTEMS recommendations for Level 1 lead to conservatively small displacement capacity. POLB recommendations are also adequate for Level 2 but not Level 1 for concrete and are overly conservative for steal and therefore, not adequate for either level except in dense and medium sands during a Level 1earthquake. POLB does not take into account different soil characteristics and has one value for all soils, which is inadequate for most cases.

Water supply reservoirs are often subject to accelerated nutrient enrichment from urban sources. Cultural eutrophication due to such enrichment requires the development of efficient management and remediation strategies to protect drinking water sources. This study investigates the effects of using nitrate as part of a management strategy to control nutrient cycling in the Occoquan Reservoir in northern Virginia, USA. A novel aspect of the study is that the reservoir is part of an indirect potable reuse system where the source of nitrate is the product water from an advanced water reclamation facility (WRF). Field and laboratory studies showed that nitrate at a concentration greater than 1 mg/L N was effective in controlling the release of phosphorus, iron, and manganese from the sediments after the depletion of oxygen from the hypolimnetic waters of the reservoir. However, when the nitrate concentration above the sediment-water interface was less than 1 mg/L N, phosphorus, iron, and manganese release from the sediments was evident. Experiments revealed that the presence of nitrate decreased sediment ammonium release, but did not completely prevent it during anoxic periods. Results also showed that changes in the effective depth (ED) value along the length of the reservoir promoted higher denitrification rates in the upper reaches of the reservoir, thereby decreasing the downstream transport of nitrate. During periods of hypolimnetic anoxia, a nitrate-N input from the WRF of at least 10 mg/L N is needed to maintain an oxidized environment above the sediment-water interface. Therefore, decreasing the nitrate input to the reservoir will likely result in the deterioration of the surface water quality in the reservoir. Finally, the ED concept was proven to be an effective method to simulate different segments of the reservoir in laboratory-scaled experiments. Similarities between the field and laboratory results suggests that the environment that existed in the waters of the reservoir was closely replicated in the experimental setup, and provides confidence that laboratory results can be extrapolated to natural reservoir conditions. / Ph. D.

No / We present a novel self-supervised framework for monocular image depth learning and confidence estimation. Our framework reduces the amount of ground truth annotation data required for training Convolutional Neural Networks (CNNs), which is often a challenging problem for the fast deployment of CNNs in many computer vision tasks. Our DepthNet adopts a novel fully differential patch-based cost function through the Zero-Mean Normalized Cross Correlation (ZNCC) to take multi-scale patches as matching and learning strategies. This approach greatly increases the accuracy and robustness of the depth learning. Whilst the proposed patch-based cost function naturally provides a 0-to-1 confidence, it is then used to self-supervise the training of a parallel network for confidence map learning and estimation by exploiting the fact that ZNCC is a normalized measure of similarity which can be approximated as the confidence of the depth estimation. Therefore, the proposed corresponding confidence map learning and estimation operate in a self-supervised manner and is a parallel network to the DepthNet. Evaluation on the KITTI depth prediction evaluation dataset and Make3D dataset show that our method outperforms the state-of-the-art results.

<p> Monthly median and annual mean values of optical depth and the ratio of diffuse to direct solar radiation for 1981-1983 were calculated using integrated values of global and diffuse radiation and calculations of precipitable water, under cloudless conditions. Results indicate that El Chichon's volcanic dust cloud has affected turbidity over southern Ontario. This is reflected in an increase in optical depth and the ratio of diffuse to direct solar radiation. </p> / Thesis / Bachelor of Arts (BA)

In this thesis, findings in three areas of computer vision estimation are presented. First, an improvement to the Kanade-Lucas-Tomasi (KLT) feature tracking algorithm is presented in which gyroscope data is incorporated to compensate for camera rotation. This improved algorithm is then compared with the original algorithm and shown to be more effective at tracking features in the presence of large rotational motion. Next, a deep neural network approach to depth estimation is presented. Equations are derived relating camera and feature motion to depth. The information necessary for depth estimation is given as inputs to a deep neural network, which is trained to predict depth across an entire scene. This deep neural network approach is shown to be effective at predicting the general structure of a scene. Finally, a method of passively estimating the position and velocity of constant velocity targets using only bearing and time-to-collision measurements is presented. This method is paired with a path planner to avoid tracked targets. Results are given to show the effectiveness of the method at avoiding collision while maneuvering as little as possible.

In appreciation of High-Performance Computing, modern scientific simulations are scaling into millions and even billions of grid points. As we enter the exa-scale, new strategies are required for visualization and analysis. While Image-Based Rendering (IBR) has emerged as a viable solution to the asymmetry between data size and its storage and required rendering power, it is limited in its 2D image portrayal of 3D spatial objects. This work describes a novel technique to capture, represent, and render depth information in the context of 3D IBR. We tested the value of displacement by displacement map, shading by normal, and image angle interval with our technique. We ran an online user study of 60 participants to evaluate the value of adding depth information back to Image-Based Rendering and found significant benefits. / Master of Science / In scientific research, data visualization is important for better understanding data. Modern experiments and simulations are expanding rapidly in scale, and there will come a day when rendering the entire 3D geometry becomes impossible resource-wise. Cinema was proposed as an image-Based solution to this problem, where the model was represented by an interpolated series of images. However, using flat images cannot fully express the 3D characteristics of a data. Therefore, in this work, we try to improve the depth portrayal of the images by protruding the pixels and applying shading. We show the results of a user study conducted with 60 participants on the effect of pixel protrusion, shading, and varying the number of images representing the object. Results show that this method would be useful for 3D scientific visualizations. The resulting object almost accurately resembles the 3D object.

In western North America, winterfat (Krascheninnikovia lanata (Pursh) A. Meeuse & Smit) is a valuable protein-rich subshrub whose restoration has been limited by poor seed flowability and low rates of seedling establishment. Seed flowability can be limited by a dense covering of hairs on winterfat fruits that can cause them to clog in mechanized equipment. Seedling establishment can be limited by premature germination of fall-sown seeds that can cause over-winter seedling mortality from freezing, pathogen attack, and winter drought. Seed coatings may provide a way to overcome both of these barriers to winterfat restoration. Coatings can compress hairs against the fruit and improve seed flowability, and a hydrophobic polymer within seed coatings can repel water and delay germination of fall-sown seeds until spring, when winter hazards have subsided, and conditions are more conducive to seedling establishment. With the advent of this technology, there is a need to establish cultural practices, such as optimal planting depth, for coated winterfat fruits. In chapter 1 of this thesis, we evaluated the influence of planting depth, seed coatings, and their interactions on winterfat seedling emergence under laboratory and field conditions. We predicted that seedling emergence would be greatest from shallow planting depths, and that coatings would not affect emergence. Results generally supported our hypothesis, with seedling emergence being highest from surface-sown and shallow-planted seeds for both non-coated (control) and coated winterfat fruits in laboratory and field conditions. Emergence from surface-sown seeds was more than two-fold greater than from the deepest planting depth (12.7 mm). Seed coatings improved emergence of surface-sown seeds compared to the control by 52 – 168% in the laboratory but had no effect in the field. As predicted, emergence was similar between coated and non-coated fruits when sown below the soil surface in both laboratory and field conditions. These results suggest that seed coatings may improve winterfat restoration success by improving flowability without inhibiting emergence, allowing the species to be used in more seeding projects. Winterfat seed coatings may be improved with the use of a hydrophobic polymer to delay germination of fall-sown seeds until spring. In chapter 2 of this thesis, we compared seedling emergence from non-coated seeds, calcium carbonate coated seeds (blank-coated), and seeds coated with calcium carbonate plus an exterior hydrophobic coating. We counted the number of live seedlings and those that had died after emerging, and calculated mortality percentages for each treatment. We hypothesized that emergence would be greatest from hydrophobic-coated seeds, and the results supported our hypothesis. Seedling emergence from hydrophobic-coated seeds was three-fold greater than the control, and five-fold greater than blank-coated seeds. Mortality percentages were highest for the control, lower for blank-coated seeds, and lowest for hydrophobic-coated seeds. Thus, hydrophobic seed coatings can improve winterfat seedling emergence, and so could be instrumental restoring this valuable species to degraded rangelands.

Much about visual spatial attention has been learned from studying how observers respond to two-dimensional stimuli. Less is known about how attention varies along the depth axis. Most of the work on the effect of depth on spatial attention manipulated binocular disparity defined depth, and it is less clear how monocular depth cues affect spatial attention. This thesis investigates the effect of target distance on peripheral detection in a virtual three-dimensional environment that simulated distance using pictorial and motion cues. Participants followed a lead car at a constant distance actively or passively, while travelling along a straight trajectory. The horizontal distribution of attention was measured using a peripheral target detection task. Both car-following and peripheral detection were tested alone under focussed attention, and simultaneously under divided attention. Chapter 2 evaluated the effect of target distance and eccentricity on peripheral detection. Experiment 1 found an overall near advantage that increased at larger eccentricities. Experiment 2 examined the effect of anticipation on target detection and found that equating anticipation across distances drastically reduced the effect of distance in reaction time, but did not affect accuracy. Experiments 3 and 4 examined the relative contributions of pictorial cues on the effect of target distance and found that the background texture that surrounded the targets could explain the main effect of distance but could not fully account for the interaction between distance and eccentricity. Chapter 3 extended the findings of Chapter 2 and found that the effect of distance on peripheral detection in our conditions was non-monotonic and did not depend on fixation distance. Across chapters, dividing attention between the central car-following and peripheral target detection tasks consistently resulted in costs for car-following, but not for peripheral detection. This work has implications for understanding spatial attention and design of advanced driver assistance systems. / Dissertation / Doctor of Science (PhD) / Our visual world is complex and dynamic, and spatial attention enables us to focus on certain relevant locations of our world. However, much of what we know about spatial attention has been studied in the context of a two-dimensional plane, and less is known about how it varies in the third dimension: depth. This thesis aims to better understand how spatial attention is affected by depth in a virtual three-dimensional environment, particularly in a driving context. Generally, driving was simulated using a car-following task, spatial attention was measured in a task that required detecting targets appearing at different depths indicated by cues perceivable with one eye. The results of this work add to the literature that suggests that spatial attention is affected by depth and contributes to our understanding of how attention may be allocated in space. Additionally, this thesis may have implications for the design of in-car warning systems.