VTechWorks Repository :: Browsing by Author "Abbott, A. Lynn"

Browsing by Author "Abbott, A. Lynn"

Now showing 1 - 20 of 309

3-D Point Cloud Generation from Rigid and Flexible Stereo Vision Systems
Short, Nathaniel Jackson (Virginia Tech, 2009-12-04)
When considering the operation of an Unmanned Aerial Vehicle (UAV) or an Unmanned Ground Vehicle (UGV), such problems as landing site estimation or robot path planning become a concern. Deciding if an area of terrain has a level enough slope and a wide enough area to land a Vertical Take Off and Landing (VTOL) UAV or if an area of terrain is traversable by a ground robot is reliant on data gathered from sensors, such as cameras. 3-D models, which can be built from data extracted from digital cameras, can help facilitate decision making for such tasks by providing a virtual model of the surrounding environment the system is in. A stereo vision system utilizes two or more cameras, which capture images of a scene from two or more viewpoints, to create 3-D point clouds. A point cloud is a set of un-gridded 3-D points corresponding to a 2-D image, and is used to build gridded surface models. Designing a stereo system for distant terrain modeling requires an extended baseline, or distance between the two cameras, in order to obtain a reasonable depth resolution. As the width of the baseline increases, so does the flexibility of the system, causing the orientation of the cameras to deviate from their original state. A set of tools have been developed to generate 3-D point clouds from rigid and flexible stereo systems, along with a method for applying corrections to a flexible system to regain distance accuracy in a flexible system.
3D Deep Learning for Object-Centric Geometric Perception
Li, Xiaolong (Virginia Tech, 2022-06-30)
Object-centric geometric perception aims at extracting the geometric attributes of 3D objects. These attributes include shape, pose, and motion of the target objects, which enable fine-grained object-level understanding for various tasks in graphics, computer vision, and robotics. With the growth of 3D geometry data and 3D deep learning methods, it becomes more and more likely to achieve such tasks directly using 3D input data. Among different 3D representations, a 3D point cloud is a simple, common, and memory-efficient representation that could be directly retrieved from multi-view images, depth scans, or LiDAR range images. Different challenges exist in achieving object-centric geometric perception, such as achieving a fine-grained geometric understanding of common articulated objects with multiple rigid parts, learning disentangled shape and pose representations with fewer labels, or tackling dynamic and sequential geometric input in an end-to-end fashion. Here we identify and solve these challenges from a 3D deep learning perspective by designing effective and generalizable 3D representations, architectures, and pipelines. We propose the first deep pose estimation for common articulated objects by designing a novel hierarchical invariant representation. To push the boundary of 6D pose estimation for common rigid objects, a simple yet effective self-supervised framework is designed to handle unlabeled partial segmented scans. We further contribute a novel 4D convolutional neural network called PointMotionNet to learn spatio-temporal features for 3D point cloud sequences. All these works advance the domain of object-centric geometric perception from a unique 3D deep learning perspective.
Action Recognition with Knowledge Transfer
Choi, Jin-Woo (Virginia Tech, 2021-01-07)
Recent progress on deep neural networks has shown remarkable action recognition performance from videos. The remarkable performance is often achieved by transfer learning: training a model on a large-scale labeled dataset (source) and then fine-tuning the model on the small-scale labeled datasets (targets). However, existing action recognition models do not always generalize well on new tasks or datasets because of the following two reasons. i) Current action recognition datasets have a spurious correlation between action types and background scene types. The models trained on these datasets are biased towards the scene instead of focusing on the actual action. This scene bias leads to poor generalization performance. ii) Directly testing the model trained on the source data on the target data leads to poor performance as the source, and target distributions are different. Fine-tuning the model on the target data can mitigate this issue. However, manual labeling small- scale target videos is labor-intensive. In this dissertation, I propose solutions to these two problems. For the first problem, I propose to learn scene-invariant action representations to mitigate the scene bias in action recognition models. Specifically, I augment the standard cross-entropy loss for action classification with 1) an adversarial loss for the scene types and 2) a human mask confusion loss for videos where the human actors are invisible. These two losses encourage learning representations unsuitable for predicting 1) the correct scene types and 2) the correct action types when there is no evidence. I validate the efficacy of the proposed method by transfer learning experiments. I trans- fer the pre-trained model to three different tasks, including action classification, temporal action localization, and spatio-temporal action detection. The results show consistent improvement over the baselines for every task and dataset. I formulate human action recognition as an unsupervised domain adaptation (UDA) problem to handle the second problem. In the UDA setting, we have many labeled videos as source data and unlabeled videos as target data. We can use already exist- ing labeled video datasets as source data in this setting. The task is to align the source and target feature distributions so that the learned model can generalize well on the target data. I propose 1) aligning the more important temporal part of each video and 2) encouraging the model to focus on action, not the background scene, to learn domain-invariant action representations. The proposed method is simple and intuitive while achieving state-of-the-art performance without training on a lot of labeled target videos. I relax the unsupervised target data setting to a sparsely labeled target data setting. Then I explore the semi-supervised video action recognition, where we have a lot of labeled videos as source data and sparsely labeled videos as target data. The semi-supervised setting is practical as sometimes we can afford a little bit of cost for labeling target data. I propose multiple video data augmentation methods to inject photometric, geometric, temporal, and scene invariances to the action recognition model in this setting. The resulting method shows favorable performance on the public benchmarks.
An Adaptive Computer Vision Technique for Estimating the Biomass and Density of Loblolly Pine Plantations using Digital Orthophotography and LiDAR Imagery
Bortolot, Zachary Jared (Virginia Tech, 2004-04-23)
Forests have been proposed as a means of reducing atmospheric carbon dioxide levels due to their ability to store carbon as biomass. To quantify the amount of atmospheric carbon sequestered by forests, biomass and density estimates are often needed. This study develops, implements, and tests an individual tree-based algorithm for obtaining forest density and biomass using orthophotographs and small footprint LiDAR imagery. It was designed to work with a range of forests and image types without modification, which is accomplished by using generic properties of trees found in many types of images. Multiple parameters are employed to determine how these generic properties are used. To set these parameters, training data is used in conjunction with an optimization algorithm (a modified Nelder-Mead simplex algorithm or a genetic algorithm). The training data consist of small images in which density and biomass are known. A first test of this technique was performed using 25 circular plots (radius = 15 m) placed in young pine plantations in central Virginia, together with false color othophotograph (spatial resolution = 0.5 m) or small footprint LiDAR (interpolated to 0.5 m) imagery. The highest density prediction accuracies (r2 up to 0.88, RMSE as low as 83 trees / ha) were found for runs where photointerpreted densities were used for training and testing. For tests run using density measurements made on the ground, accuracies were consistency higher for orthophotograph-based results than for LiDAR-based results, and were higher for trees with DBH ≥10cm than for trees with DBH ≥7 cm. Biomass estimates obtained by the algorithm using LiDAR imagery had a lower RMSE (as low as 15.6 t / ha) than most comparable studies. The correlations between the actual and predicted values (r2 up to 0.64) were lower than comparable studies, but were generally highly significant (p ≤ 0.05 or 0.01). In all runs there was no obvious relationship between accuracy and the amount of training data used, but the algorithm was sensitive to which training and testing data were selected. Methods were evaluated for combining predictions made using different parameter sets obtained after training using identical data. It was found that averaging the predictions produced improved results. After training using density estimates from the human photointerpreter, 89% of the trees located by the algorithm corresponded to trees found by the human photointerpreter. A comparison of the two optimization techniques found them to be comparable in speed and effectiveness.
Addressing Occlusion in Panoptic Segmentation
Sarkaar, Ajit Bhikamsingh (Virginia Tech, 2021-01-20)
Visual recognition tasks have witnessed vast improvements in performance since the advent of deep learning. Despite the gains in performance, image understanding algorithms are still not completely robust to partial occlusion. In this work, we propose a novel object classification method based on compositional modeling and explore its effect in the context of the newly introduced panoptic segmentation task. The panoptic segmentation task combines both semantic and instance segmentation to perform labelling of the entire image. The novel classification method replaces the object detection pipeline in UPSNet, a Mask R-CNN based design for panoptic segmentation. We also discuss an issue with the segmentation mask prediction of Mask R-CNN that affects overlapping instances. We perform extensive experiments and showcase results on the complex COCO and Cityscapes datasets. The novel classification method shows promising results for object classification on occluded instances in complex scenes.
Adversarial Learning based framework for Anomaly Detection in the context of Unmanned Aerial Systems
Bhaskar, Sandhya (Virginia Tech, 2020-06-18)
Anomaly detection aims to identify the data samples that do not conform to a known normal (regular) behavior. As the definition of an anomaly is often ambiguous, unsupervised and semi-supervised deep learning (DL) algorithms that primarily use unlabeled datasets to model normal (regular) behaviors, are popularly studied in this context. The unmanned aerial system (UAS) can use contextual anomaly detection algorithms to identify interesting objects of concern in applications like search and rescue, disaster management, public security etc. This thesis presents a novel multi-stage framework that supports detection of frames with unknown anomalies, localization of anomalies in the detected frames, and validation of detected frames for incremental semi-supervised learning, with the help of a human operator. The proposed architecture is tested on two new datasets collected for a UAV-based system. In order to detect and localize anomalies, it is important to both model the normal data distribution accurately as well as formulate powerful discriminant (anomaly scoring) techniques. We implement a generative adversarial network (GAN)-based anomaly detection architecture to study the effect of loss terms and regularization on the modeling of normal (regular) data and arrive at the most effective anomaly scoring method for the given application. Following this, we use incremental semi-supervised learning techniques that utilize a small set of labeled data (obtained through validation from a human operator), with large unlabeled datasets to improve the knowledge-base of the anomaly detection system.
Aerial high-throughput phenotyping of peanut leaf area index and lateral growth
Sarkar, Sayantan; Cazenave, Alexandre-Brice; Oakes, Joseph C.; McCall, David S.; Thomason, Wade E.; Abbott, A. Lynn; Balota, Maria (Springer Nature, 2021-11-04)
Leaf area index (LAI) is the ratio of the total one-sided leaf area to the ground area, whereas lateral growth (LG) is the measure of canopy expansion. They are indicators for light capture, plant growth, and yield. Although LAI and LG can be directly measured, this is time consuming. Healthy leaves absorb in the blue and red, and reflect in the green regions of the electromagnetic spectrum. Aerial high-throughput phenotyping (HTP) may enable rapid acquisition of LAI and LG from leaf reflectance in these regions. In this paper, we report novel models to estimate peanut (Arachis hypogaea L.) LAI and LG from vegetation indices (VIs) derived relatively fast and inexpensively from the red, green, and blue (RGB) leaf reflectance collected with an unmanned aerial vehicle (UAV). In addition, we evaluate the models’ suitability to identify phenotypic variation for LAI and LG and predict pod yield from early season estimated LAI and LG. The study included 18 peanut genotypes for model training in 2017, and 8 genotypes for model validation in 2019. The VIs included the blue green index (BGI), red-green ratio (RGR), normalized plant pigment ratio (NPPR), normalized green red difference index (NGRDI), normalized chlorophyll pigment index (NCPI), and plant pigment ratio (PPR). The models used multiple linear and artificial neural network (ANN) regression, and their predictive accuracy ranged from 84 to 97%, depending on the VIs combinations used in the models. The results concluded that the new models were time- and cost-effective for estimation of LAI and LG, and accessible for use in phenotypic selection of peanuts with desirable LAI, LG and pod yield.
Analysis and Reduction of Moire Patterns in Scanned Halftone Pictures
Liu, Xiangdong (Virginia Tech, 1996-05-01)
In this dissertation we provide a comprehensive theory for the formation of a moire pattern in a sampled halftone image. We explore techniques for restoring a sampled halftone image with a moire pattern and techniques for preventing a moire pattern when a halftone picture is scanned. Specifically, we study the frequency, phase, and spatial geometry of a moire pattern. We observe and explain the half period phase reversal phenomenon that a moire pattern may exhibit. As a case study, we examine the moire patterns generated by a commercial scanner. We propose three restoration methods, including a notch filtering method, a simulation method, and a relaxation method. We also describe a moire prevention method, the partial inverse Fourier transform method. Finally, we propose a research agenda for further investigation.
An Analysis of EcoRouting Using a Variable Acceleration Rate Synthesis Model
Warpe, Hrusheekesh Sunil (Virginia Tech, 2017-08-07)
Automotive manufacturers are facing increasing pressure from legislative bodies and consumers to reduce fuel consumption and greenhouse gas emissions of vehicles. This has led to many automotive manufacturers starting production of Plug-in Hybrid Electric Vehicles (PHEV's) and Battery Electric Vehicles (BEV's). Another method that helps to reduce the environmental effect of transportation is EcoRouting. The standard Global Positioning System (GPS) navigation offers route alternatives between user specified origin and destination. This technology provides multiple routes to the user and focuses on reducing the travel time to reach to the destination. EcoRouting is the method to determine a route that minimizes vehicle energy consumption, unlike traditional routing methods that minimize travel time. An EcoRouting system has been developed as a part of this thesis that takes in information such as speed limits, the number of stop lights, and the road grade to calculate the energy consumption of a vehicle along a route. A synthesis methodology is introduced that takes into consideration the distance between the origin and destination, the acceleration rate of the vehicle, cruise speed and jerk rate as inputs to simulate driver behavior on a given route. A new approach is presented in this thesis that weighs the energy consumption for different routes and chooses the route with the least energy consumption, subject to a constraint on travel time. A cost function for quantifying the effect of travel time is introduced that assists in choosing the EcoRoute with an acceptable limit on the travel time required to reach the destination. The analysis of the EcoRouting system with minimum number of conditional stops and maximum number of conditional stops is done in this thesis. The effect on energy consumption with the presence and absence of road-grade information along a route is also studied. A sensitivity study is performed to observe the change in energy consumption of the vehicle with a change in acceleration rates and road grade. Three routing scenarios are presented in this thesis to demonstrate the functionality of EcoRouting. The EcoRouting model presented in this thesis is also validated against an external EcoRouting research paper and the energy consumption along three routes is calculated. The EcoRoute solution is found to vary with the information given to the variable acceleration rate model. The synthesis and the results that are obtained show that parameters such as acceleration, deceleration, and road grade affect the overall energy consumption of a vehicle and are helpful in determining the EcoRoute.
Analyzing and Classifying Neural Dynamics from Intracranial Electroencephalography Signals in Brain-Computer Interface Applications
Nagabushan, Naresh (Virginia Tech, 2019-06-14)
Brain-Computer Interfaces (BCIs) that rely on motor imagery currently allow subjects to control quad-copters, robotic arms, and computer cursors. Recent advancements have been made possible because of breakthroughs in fields such as electrical engineering, computer science, and neuroscience. Currently, most real-time BCIs use hand-crafted feature extractors, feature selectors, and classification algorithms. In this work, we explore the different classification algorithms currently used in electroencephalographic (EEG) signal classification and assess their performance on intracranial EEG (iEEG) data. We first discuss the motor imagery task employed using iEEG signals and find features that clearly distinguish between different classes. Second, we compare the different state-of-the-art classifiers used in EEG BCIs in terms of their error rate, computational requirements, and feature interpret-ability. Next, we show the effectiveness of these classifiers in iEEG BCIs and last, show that our new classification algorithm that is designed to use spatial, spectral, and temporal information reaches performance comparable to other state-of-the-art classifiers while also allowing increased feature interpret-ability.
An Antenna Specific Site Modeling Tool for Interactive Computation of Coverage Regions for Indoor Wireless Communication
Bhat, Nitin (Virginia Tech, 1998-03-02)
A goal of indoor wireless communication is to strategically place RF base stations to obtain optimum signal coverage at the lowest cost and power. Traditionally, transceiver locations have been selected by human experts who rely on experience and heuristics to obtain a near-optimum placement. Current methods depend on involved on-site communication measurements and crude statistical modeling of the obtained data which is time consuming and prohibitive in cost. Given the inherent variability of the indoor environment, such a method often yields poor efficiency. As an example, it is possible that more power than required or extra number of transceivers were used. This thesis describes an interactive software system that can be used to aid transceiver placement. The tool is easy to use and is targeted at users who are not experts in wireless communication system design. Once the transceiver locations are selected by the user within a graphical floor plan, the system uses simple path-loss models to predict coverage regions for each transceiver. The coverage regions are highlighted to indicate expected coverage. Earlier work assumed isotropic transceivers and had limited directional transmitter support. This thesis describes how the tool has been enhanced to support a wide range of 3D antenna patterns as encountered in practical situations. The tool has also been expanded to accommodate more partition types and to report area of coverage. The resulting system is expected to be very useful in the practical deployment of indoor wireless systems.
Application of Computer Vision Techniques for Railroad Inspection using UAVs
Harekoppa, Pooja Puttaswamygowda (Virginia Tech, 2016-08-16)
The task of railroad inspection is a tedious one. It requires a lot of skilled experts and long hours of frequent on-field inspection. Automated ground equipment systems that have been developed to address this problem have the drawback of blocking the rail service during inspection process. As an alternative, using aerial imagery from a UAV, Computer Vision and Machine Learning based techniques were developed in this thesis to analyze two kinds of defects on the rail tracks. The defects targeted were missing spikes on tie plates and cracks on ties. In order to perform this inspection, the rail region was identified in the image and then the tie plate and tie regions on the track were detected. These steps were performed using morphological operations, filtering and intensity analysis. Once the tie plate was localized, the regions of interest on the plate were used to train a machine learning model to recognize missing spikes. Classification using SVM resulted in an accuracy of around 96% and varied greatly based on the tie plate illumination and ROI alignment for Lampasas and Chickasha subdivision datasets. Also, many other different classifiers were used for training and testing and an ensemble method with majority vote scheme was also explored for classification. The second category of learning model used was a multi-layered neural network. The major drawback of this method was, it required a lot of images for training. However, it performed better than feature based classifiers with availability of larger training dataset. As a second kind of defect, tie conditions were analyzed. From the localized tie region, the tie cracks were detected using thresholding and morphological operations. A machine learning classifier was developed to predict the condition of a tie based on training examples of images with extracted features. The multi-class classification accuracy obtained was around 83% and there were no misclassifications seen between two extreme classes of tie condition on the test data.
Appling Machine and Statistical Learning Techniques to Intelligent Transport Systems: Bottleneck Identification and Prediction, Dynamic Travel Time Prediction, Driver Run-Stop Behavior Modeling, and Autonomous Vehicle Control at Intersections
Elhenawy, Mohammed Mamdouh Zakaria (Virginia Tech, 2015-06-30)
In this dissertation, new algorithms that address three traffic problems of major importance are developed. First automatic identification and prediction algorithms are developed to identify and predict the occurrence of traffic congestion. The identification algorithms concoct a model to identify speed thresholds by exploiting historical spatiotemporal speed matrices. We employ the speed model to define a cutoff speed separating free-flow from congested traffic. We further enhance our algorithm by utilizing weather and visibility data. To our knowledge, we are the first to include weather and visibility variables in formulating an automatic congestion identification model. We also approach the congestion prediction problem by adopting an algorithm which employs Adaptive Boosting machine learning classifiers again something novel that has not been done previously. The algorithm is promising where it resulted in a true positive rate slightly higher than 0.99 and false positive rate less than 0.001. We next address the issue of travel time modeling. We propose algorithms to model travel time using various machine learning and statistical learning techniques. We obtain travel time models by employing the historical spatiotemporal speed matrices in conjunction with our algorithms. The algorithms yield pertinent information regarding travel time reliability and prediction of travel times. Our proposed algorithms give better predictions compared to the state of practice algorithms. Finally we consider driver safety at signalized intersections and uncontrolled intersections in a connected vehicles environment. For signalized intersections, we exploit datasets collected from four controlled experiments to model the stop-run behavior of the driver at the onset of the yellow indicator for various roadway surface conditions and multiple vehicle types. We further propose a new variable (predictor) related to driver aggressiveness which we estimate by monitoring how drivers respond to yellow indications. The performance of the stop-run models shows improvements after adding the new aggressiveness predictor. The proposed models are practical and easy to implement in advanced driver assistance systems. For uncontrolled intersections, we present a game theory based algorithm that models the intersection as a chicken game to solve the conflicts between vehicles crossing the intersection. The simulation results show a 49% saving in travel time on average relative to a stop control when the vehicles obey the Nash equilibrium of the game.
Arc Path Collision Avoidance Algorithm for Autonomous Ground Vehicles
Naik, Ankur (Virginia Tech, 2005-12-15)
Presented in this thesis is a collision avoidance algorithm designed around an arc path model. The algorithm was designed for use on Virginia Tech robots entered in the 2003 and 2004 Intelligent Ground Vehicle Competition (IGVC) and on our 2004 entry into the DARPA Grand Challenge. The arc path model was used because of the simplicity of the calculations and because it can accurately represent the base kinematics for Ackerman or differentially steered vehicles. Clothoid curves have been used in the past to create smooth paths with continuously varying curvature, but clothoids are computationally intensive. The circular arc algorithm proposed here is designed with simplicity and versatility in mind. It is readily adaptable to ground vehicles of any size and shape. The algorithm is also designed to run with minimal tuning. The algorithm can be used as a stand alone reactive collision avoidance algorithm in simple scenarios, but it can be better optimized for speed and safety when guided by a global path planner. A complete navigation architecture is presented as an example of how obstacle avoidance can be incorporated in the algorithm.
The Art of Deep Connection - Towards Natural and Pragmatic Conversational Agent Interactions
Ray, Arijit (Virginia Tech, 2017-07-12)
As research in Artificial Intelligence (AI) advances, it is crucial to focus on having seamless communication between humans and machines in order to effectively accomplish tasks. Smooth human-machine communication requires the machine to be sensible and human-like while interacting with humans, while simultaneously being capable of extracting the maximum information it needs to accomplish the desired task. Since a lot of the tasks required to be solved by machines today involve the understanding of images, training machines to have human-like and effective image-grounded conversations with humans is one important step towards achieving this goal. Although we now have agents that can answer questions asked for images, they are prone to failure from confusing input, and cannot ask clarification questions, in turn, to extract the desired information from humans. Hence, as a first step, we direct our efforts towards making Visual Question Answering agents human-like by making them resilient to confusing inputs that otherwise do not confuse humans. Not only is it crucial for a machine to answer questions reasonably, it should also know how to ask questions sequentially to extract the desired information it needs from a human. Hence, we introduce a novel game called the Visual 20 Questions Game, where a machine tries to figure out a secret image a human has picked by having a natural language conversation with the human. Using deep learning techniques like recurrent neural networks and sequence-to-sequence learning, we demonstrate scalable and reasonable performances on both the tasks.
Assessment of Psychophysiological Characteristics of Drivers Using Heart Rate from SHRP2 Face Video Data
Sarkar, Abhijit; Doerzaph, Zachary R.; Abbott, A. Lynn (2014-08-25)
The goal is to

Extract heart rate from face video

Understand the behavior of driver, e.g. cognitive load, panic attack, drowsiness, DUI

Develop automatic video reduction technique

Devise a tool for future
Automated Detection of Surface Defects on Barked Hardwood Logs and Stems Using 3-D Laser Scanned Data
Thomas, Liya (Virginia Tech, 2006-09-08)
This dissertation presents an automated detection algorithm that identifies severe external defects on the surfaces of barked hardwood logs and stems. The defects detected are at least 0.5 inch in height and at least 3 inches in diameter, which are severe, medium to large in size, and have external surface rises. Hundreds of real log defect samples were measured, photographed, and categorized to summarize the main defect features and to build a defect knowledge base. Three-dimensional laser-scanned range data capture the external log shapes and portray bark pattern, defective knobs, and depressions. The log data are extremely noisy, have missing data, and include severe outliers induced by loose bark that dangles from the log trunk. Because the circle model is nonlinear and presents both additive and non-additive errors, a new robust generalized M-estimator has been developed that is different from the ones proposed in the statistical literature for linear regression. Circle fitting is performed by standardizing the residuals via scale estimates calculated by means of projection statistics and incorporated in the Huber objective function to bound the influence of the outliers in the estimates. The projection statistics are based on 2-D radial-vector coordinates instead of the row vectors of the Jacobian matrix as proposed in the statistical literature dealing with linear regression. This approach proves effective in that it makes the GM-estimator to be influence bounded and thereby, robust against outliers. Severe defects are identified through the analysis of 3-D log data using decision rules obtained from analyzing the knowledge base. Contour curves are generated from radial distances, which are determined by robust 2-D circle fitting to the log-data cross sections. The algorithm detected 63 from a total of 68 severe defects. There were 10 non-defective regions falsely identified as defects. When these were calculated as areas, the algorithm locates 97.6% of the defect area, and falsely identifies 1.5% of the total clear area as defective.
Automated Landing Site Evaluation for Semi-Autonomous Unmanned Aerial Vehicles
Klomparens, Dylan (Virginia Tech, 2008-08-20)
A system is described for identifying obstacle-free landing sites for a vertical-takeoff-and-landing (VTOL) semi-autonomous unmanned aerial vehicle (UAV) from point cloud data obtained from a stereo vision system. The relatively inexpensive, commercially available Bumblebee stereo vision camera was selected for this study. A "point cloud viewer" computer program was written to analyze point cloud data obtained from 2D images transmitted from the UAV to a remote ground station. The program divides the point cloud data into segments, identifies the best-fit plane through the data for each segment, and performs an independent analysis on each segment to assess the feasibility of landing in that area. The program also rapidly presents the stereo vision information and analysis to the remote mission supervisor who can make quick, reliable decisions about where to safely land the UAV. The features of the program and the methods used to identify suitable landing sites are presented in this thesis. Also presented are the results of a user study that compares the abilities of humans and computer-supported point cloud analysis in certain aspects of landing site assessment. The study demonstrates that the computer-supported evaluation of potential landing sites provides an immense benefit to the UAV supervisor.
Automated Mapping of Typical Cropland Strips in the North China Plain Using Small Unmanned Aircraft Systems (sUAS) Photogrammetry
Zhang, Jianyong; Zhao, Yanling; Abbott, A. Lynn; Wynne, Randolph H.; Hu, Zhenqi; Zou, Yuzhu; Tian, Shuaishuai (MDPI, 2019-10-10)
Accurate mapping of agricultural fields is needed for many purposes, including irrigation decisions and cadastral management. This paper is concerned with the automated mapping of cropland strips that are common in the North China Plain. These strips are commonly 3–8 m in width and 50–300 m in length, and are separated by small ridges that assist with irrigation. Conventional surveying methods are labor-intensive and time-consuming for this application, and only limited performance is possible with very high resolution satellite images. Small Unmanned Aircraft System (sUAS) images could provide an alternative approach to ridge detection and strip mapping. This paper presents a novel method for detecting cropland strips, utilizing centimeter spatial resolution imagery captured by sUAS flying at low altitude (60 m). Using digital surface models (DSM) and ortho-rectified imagery from sUAS data, this method extracts candidate ridge locations by surface roughness segmentation in combination with geometric constraints. This method then exploits vegetation removal and morphological operations to refine candidate ridge elements, leading to polyline-based representations of cropland strip boundaries. This procedure has been tested using sUAS data from four typical cropland plots located approximately 60 km west of Jinan, China. The plots contained early winter wheat. The results indicated an ability to detect ridges with comparatively high recall and precision (96.8% and 95.4%, respectively). Cropland strips were extracted with over 98.9% agreement relative to ground truth, with kappa coefficients over 97.4%. To our knowledge, this method is the first to attempt cropland strip mapping using centimeter spatial resolution sUAS images. These results have demonstrated that sUAS mapping is a viable approach for data collection to assist in agricultural land management in the North China Plain.
Automatic Detection of Elongated Objects in X-Ray Images of Luggage
Liu, Wenye III (Virginia Tech, 1997-09-05)
This thesis presents a part of the research work at Virginia Tech on developing a prototype automatic luggage scanner for explosive detection, and it deals with the automatic detection of elongated objects (detonators) in x-ray images using matched filtering, the Hough transform, and information fusion techniques. A sophisticated algorithm has been developed for detonator detection in x-ray images, and computer software utilizing this algorithm was programmed to implement the detection on both UNIX and PC platforms. A variety of template matching techniques were evaluated, and the filtering parameters (template size, template model, thresholding value, etc.) were optimized. A variation of matched filtering was found to be reasonably effective, while a Gabor-filtering method was found not to be suitable for this problem. The developed software for both single orientations and multiple orientations was tested on x-ray images generated on AS&E and Fiscan inspection systems, and was found to work well for a variety of images. The effects of object overlapping, luggage position on the conveyor, and detonator orientation variation were also investigated using the single-orientation algorithm. It was found that the effectiveness of the software depended on the extent of overlapping as well as on the objects the detonator overlapped. The software was found to work well regardless of the position of the luggage bag on the conveyor, and it was able to tolerate a moderate amount of orientation change.

Browsing by Author "Abbott, A. Lynn"

Results Per Page

Sort Options