Indiana University-Purdue University Indianapolis (IUPUI) / Depth estimation is increasingly becoming more important in computer vision. The requirement for autonomous systems to gauge their surroundings is of the utmost importance in order to avoid obstacles, preventing damage to itself and/or other systems or people. Depth measuring/estimation systems that use multiple cameras from multiple views can be expensive and extremely complex. And as these autonomous systems decrease in size and available power, the supporting sensors required to estimate depth must also shrink in size and power consumption. This research will concentrate on a single passive method known as Depth from Defocus (DfD), which uses an in-focus and out-of-focus image to infer the depth of objects in a scene. The major contribution of this research is the introduction of a new Deep Learning (DL) architecture to process the the in-focus and out-of-focus images to produce a depth map for the scene improving both speed and performance over a range of lighting conditions. Compared to the previous state-of-the-art multi-label graph cuts algorithms applied to the synthetically blurred dataset the DfD-Net produced a 34.30% improvement in the average Normalized Root Mean Square Error (NRMSE). Similarly the DfD-Net architecture produced a 76.69% improvement in the average Normalized Mean Absolute Error (NMAE). Only the Structural Similarity Index (SSIM) had a small average decrease of 2.68% when compared to the graph cuts algorithm. This slight reduction in the SSIM value is a result of the SSIM metric penalizing images that appear to be noisy. In some instances the DfD-Net output is mottled, which is interpreted as noise by the SSIM metric. This research introduces two methods of deep learning architecture optimization. The first method employs the use of a variant of the Particle Swarm Optimization (PSO) algorithm to improve the performance of the DfD-Net architecture. The PSO algorithm was able to find a combination of the number of convolutional filters, the size of the filters, the activation layers used, the use of a batch normalization layer between filters and the size of the input image used during training to produce a network architecture that resulted in an average NRMSE that was approximately 6.25% better than the baseline DfD-Net average NRMSE. This optimized architecture also resulted in an average NMAE that was 5.25% better than the baseline DfD-Net average NMAE. Only the SSIM metric did not see a gain in performance, dropping by 0.26% when compared to the baseline DfD-Net average SSIM value. The second method illustrates the use of a Self Organizing Map clustering method to reduce the number convolutional filters in the DfD-Net to reduce the overall run time of the architecture while still retaining the network performance exhibited prior to the reduction. This method produces a reduced DfD-Net architecture that has a run time decrease of between 14.91% and 44.85% depending on the hardware architecture that is running the network. The final reduced DfD-Net resulted in a network architecture that had an overall decrease in the average NRMSE value of approximately 3.4% when compared to the baseline, unaltered DfD-Net, mean NRMSE value. The NMAE and the SSIM results for the reduced architecture were 0.65% and 0.13% below the baseline results respectively. This illustrates that reducing the network architecture complexity does not necessarily reduce the reduction in performance. Finally, this research introduced a new, real world dataset that was captured using a camera and a voltage controlled microfluidic lens to capture the visual data and a 2-D scanning LIDAR to capture the ground truth data. The visual data consists of images captured at seven different exposure times and 17 discrete voltage steps per exposure time. The objects in this dataset were divided into four repeating scene patterns in which the same surfaces were used. These scenes were located between 1.5 and 2.5 meters from the camera and LIDAR. This was done so any of the deep learning algorithms tested would see the same texture at multiple depths and multiple blurs. The DfD-Net architecture was employed in two separate tests using the real world dataset. The first test was the synthetic blurring of the real world dataset and assessing the performance of the DfD-Net trained on the Middlebury dataset. The results of the real world dataset for the scenes that were between 1.5 and 2.2 meters from the camera the DfD-Net trained on the Middlebury dataset produced an average NRMSE, NMAE and SSIM value that exceeded the test results of the DfD-Net tested on the Middlebury test set. The second test conducted was the training and testing solely on the real world dataset. Analysis of the camera and lens behavior led to an optimal lens voltage step configuration of 141 and 129. Using this configuration, training the DfD-Net resulted in an average NRMSE, NMAE and SSIM of 0.0660, 0.0517 and 0.8028 with a standard deviation of 0.0173, 0.0186 and 0.0641 respectively.

No description available.

Accurate and usable x-ray vision has long been a goal in augmented reality (AR) research and development. X-ray vision, or the ability to comprehend location and object information when such is viewed through an opaque barrier, would be imminently useful in a variety of contexts, including industrial, disaster reconnaissance, and tactical applications. In order for x-ray vision to be a useful tool for many of these applications, it would need to extend operators’ perceptual awareness of the task or environment. The effectiveness with which x-ray vision can do this is of significant research interest and is a determinant of its usefulness in an application context. In substance, then, it is crucial to evaluate the effectiveness of x-ray vision—how does information presented through x-ray vision compare to real-world information? This approach requires narrowing as x-ray vision suffers from inherent limitations, analogous to viewing an object through a window. In both cases, information is presented beyond the local context, exists past an apparently solid object, and is limited by certain conditions. Further, in both cases, the naturally suggestive use cases occur over action space distances. These distances range from 1.5 to 30 meters and represent the area in which observers might contemplate immediate visually directed actions. These actions, simple tasks with a visual antecedent, represent action potentials for x-ray vision; in effect, x-ray vision extends an operators’ awareness and ability to visualize these actions into a new context. Thus, this work seeks to answer the question “Can a real window be replaced with an AR window?” This evaluation focuses on perceived object location, investigated through a series of experiments using visually directed actions as experimental measures. This approach leverages established methodology to investigate this topic by experimentally analyzing each of several distinct variables on a continuum between real-world depth perception and fully realized x-ray vision. It was found that a real window could not be replaced with an AR window without some loss of depth perception acuity and accuracy. However, no significant difference was found between a target viewed through an opaque wall and a target viewed through a real window.

Motion in depth and/or zooming causes defocus blur. This work presents a solution to the problem of using defocus blur and optical flow information to compute depth at points that defocus when they move. We first formulate a novel algorithm which recovers defocus blur and affine parameters simultaneously. Next we formulate a novel relationship (the blur-depth relationship) between defocus blur, relative object depth and three parameters based on camera motion and intrinsic camera parameters. We can handle the situation where a single image has points which have defocused, got sharper or are focally unperturbed. Moreover, our formulation is valid regardless of whether the defocus is due to the image plane being in front of or behind the point of sharp focus.The blur-depth relationship requires a sequence of at least three images taken with the camera moving either towards or away from the object. It can be used to obtain an initial estimate of relative depth using one of several non-linear methods. We demonstrate a solution based on the Extended Kalman Filter in which the measurement equation is the blur-depth relationship. The estimate of relative depth is then used to compute an initial estimate of camera motion parameters. In order to refine depth values, the values of relative depth and camera motion are then input into a second Extended Kalman Filter in which the measurement equations are the discrete motion equations. This set of cascaded Kalman filters can be employed iteratively over a longer sequence of images in order to further refine depth. We conduct several experiments on real scenery in order to demonstrate the range of object shapes that the algorithm can handle. We show that fairly good estimates of depth can be obtained with just three images.

Depth perception has become a heavily researched area as companies and researchers are striving towards the development of self-driving cars. Self-driving cars rely on perceiving the surrounding area, which heavily depends on technology capable of providing the system with depth perception capabilities. In this paper, we explore developing a single camera (monocular) depth prediction model that is trained on panoramic depth images. Our model makes novel use of transfer learning efficient encoder models, pre-training on a larger dataset of flat depth images, and optimizing the model for use with a Jetson Nano. Additionally, we present a training and optimization framework to make developing and testing new monocular depth perception models easier and faster. While the model failed to achieve a high frame rate, the framework and models developed are a promising starting place for future work.

Retention of engineering students to graduation and career is important business for both United States (U.S.) industries and engineering education institutions alike. Industries need competent engineers dedicated to working in the field of engineering beyond graduation in order to achieve business success and national economic growth, while engineering education institutions need retention to graduation to achieve their own business goals. This dissertation took a three-pronged approach to identifying relationships between depth and specificity of engineering and response factors related to graduation and career retention of engineers. Occupational alignment, graduate school decisions, and engineering identity were evaluated for relationships with specificity or depth of discipline within engineering degrees to evaluate if increasing the depth or specificity increased the response factors. Using historical data analysis, occupational alignment and graduate school decisions were both found to be influenced by specificity of discipline. Traditional engineering disciplines were found to report the most occupational alignment after graduation, while specific engineering disciplines were more likely to attend graduate school after graduation. Additionally, for all students reporting graduate school attendance, all specificities were most likely to align their graduate degree discipline to their undergraduate degree discipline. A national survey of undergraduate engineering students revealed that engineering identity is related to depth of discipline. Students enrolled in more specific engineering curriculum, in the form of a discipline-specific major with a concentration, reported higher engineering identity. However, the discipline-specific depth of discipline followed closely behind, indicating the impact of depth of discipline is small. The largest difference in scores between the two depths of discipline was found in students' reports of a construct termed "interest". Ultimately, this dissertation found statistically significant relationships between depth and specificity of discipline and occupational alignment, graduate school decisions, and engineering identity. Though these findings are statistically significant, they were incremental, meaning depth and specificity of discipline should not be considered the main factor of influence.

No description available.

No description available.

The influence of color on size and depth perception has been explored for a century, but there is very limited research on interventions that can reduce the color illusions. This study was motivated to identify interventions and propose design guidelines for information visualization, especially where size judgment is critical. This study replicated the color size illusion and color depth illusion on an LCD monitor and it was found that yellow is the smallest and farthest color among red, yellow, green, and blue on a white background. Three types of interventions (background brightness, border color, and background grid brightness) were tested to identify the conditions that reduce the color illusions, but all of them were not statistically significant. Based on the experiment results and literature survey, design guidelines were proposed. To extend the guidelines to the bioinformatics field, design recommendations were proposed and implementation examples were illustrated. Evaluations on design implementations were evaluated by interviewing domain experts. Additionally, the relationship between the color size illusion and the color depth illusion was explored. / Master of Science

We study the problem of proving lower bounds for depth four arithmetic circuits. Depth four circuits have been receiving much attraction when it comes to recent circuit lower bound results, as a result of the series of results culminating in the fact that strong enough lower bounds for depth four circuits will imply super-polynomial lower bounds for general arithmetic circuits, and hence solve one of the most central open problems in algebraic complexity i.e a separation between the VP and VNP classes. However despite several efforts, even for general arithmetic circuits, the best known lower bound is Omega(N log N) by Baur and Strassen (1983), where N is the number of input variables. In the case of arithmetic formulas, Kalorkoti (1985) proved a lower bound that is quadratic in the number of input variables, which has not seen any improvement since then. The situation for depth three arithmetic circuits was also similar for many years, until a recent result by Kayal et. al. (2016) achieved an almost cubic lower bound that improved over the previous best quadratic bound by Shpilka and Wigderson (1999). As the main contribution of this thesis, we prove an Omega(N^1.5) lower bound on the size of a depth four circuit, for an explicit multilinear N-variate polynomial in VNP with degree d = Theta(sqrt(N)). Our approach offers a potential route to proving a super-quadratic lower bound for depth four circuits. Taking cue from the numerous successful results recently, we use the technique of the shifted partial derivatives measure to achieve the said lower bound. Particularly, we use the Dimension of Projected Shifted Partials (DPSP) measure which has been previously used in recent depth four results. Coming to the choice of the hard polynomial, we again follow the status quo and use a variant of the Nisan-Wigderson (NW) polynomial family that has proved to be very helpful over the past few years in arithmetic circuit complexity. Finally, we do a careful analysis of Shoup-Smolensky (1997) and Raz (2010) and compare their techniques to ours. We conclude that our result can potentially be used as a starting point, and techniques similar to Kayal et. al. (2016) can likely be used to strengthen our lower bound to Omega(N^2.5), for general depth four arithmetic circuits. However, unlike depth three circuits, proving a super-quadratic lower bound for depth four circuits remains a prevalent open problem for many years. Previous work like Shoup-Smolensky and Raz implied super-linear lower bounds. To the best of our knowledge, the previous best known lower bound for general depth four circuits is Omega(N^1.33).