Search Results (1 - 7 of 7 Results)

Sort By  
Sort Dir
 
Results per page  

Bettaieb, Luc AlexandreA Deep Learning Approach To Coarse Robot Localization
Master of Sciences (Engineering), Case Western Reserve University, 2017, EECS - Electrical Engineering
This thesis explores the use of deep learning for robot localization with applications in re-localizing a mislocalized robot. Seed values for a localization algorithm are assigned based on the interpretation of images. A deep neural network was trained on images acquired in and associated with named regions. In application, the neural net was used to recognize a region based on camera input. By recognizing regions from the camera, the robot can be localized grossly, and subsequently refined with existing techniques. Explorations into different deep neural network topologies and solver types are discussed. A process for gathering training data, training the classifier, and deployment through a robot operating system (ROS) package is provided.

Committee:

Wyatt Newman (Advisor); Murat Cavusoglu (Committee Member); Gregory Lee (Committee Member)

Subjects:

Computer Science; Electrical Engineering; Robotics

Keywords:

robotics; localization; deep learning; neural networks; machine learning; state estimation; robots; robot; robot operating system; ROS; AMCL; monte carlo localization; particle filter; ConvNets; convolutional neural networks

Lao, YuanweiVisual Tracking by Exploiting Observations and Correlations
Doctor of Philosophy, The Ohio State University, 2010, Electrical and Computer Engineering

When more video cameras are employed in a wide range of applications, how to understand this huge amount of video data in an automatic way becomes a urgent task. Visual tracking is such a strong tool used to abstract the high-level information, such as human activity recognition, traffic information accumulation, security event detection, etc. However, efficiency is still one major issue limiting tracking algorithms in many real-time applications. Consequently a great deal of research effort has been focused on the design of an efficient searching strategy and a discriminative measuring method for a good tracker. Under the probabilistic framework, we notice that intermediate measurements and the correlations among the multiple targets are valuable information for the generation of samples, which have not been fully utilized before. Therefore, the objective of this research is to exploit how to improve the searching efficiency by integrating these two factors into the sampling in several applications.

We first start with a single target tracking, and we propose to update the proposal distribution by dynamically incorporating the most recent measurements and generating particles sequentially, where the contextual confidence of the user on the measurement model is also involved. In addition, the matching template is divided into non-overlapping fragments, and by learning the background information, only a subset of the most discriminative target regions are dynamically selected to measure each particle, where the model update is easily embedded to handle fast appearance changes. The two parts are dynamically fused together such that the system is able to capture abrupt motions and produce a better localization of the moving target in an efficient way.

Then we extend our attention to the case of multiple targets, and we consider a special case where targets are highly correlated, demonstrating a common motion pattern with individual variations. We propose to update the sampling distribution by incorporating both the most recent observations and the correlation information. The correlation is either known a priori or learned from the previous tracking results. In this way, the observation of a single target is multiplexed statistically through mutual correlation for other targets, and the correlation serves as both a prior information to improve the efficiency and a constraint to prevent trackers from confusing or drifting. Experiments on both synthetic and real-world data verify the effectiveness of the new algorithm and demonstrate its superiority over existing methods.

Committee:

Yuan F. Zheng (Advisor); Jose B. Cruz, Jr. (Committee Member); Ashok Krishnamurthy (Committee Member); James M. Unger (Committee Member)

Subjects:

Electrical Engineering

Keywords:

Particle filter; tracking; proposal distribution; measurement confidence; Haar; occlusion; MCMC

Raghuvanshi, AnuragParticle filter with Hyperbolic Measurements and Geometry Constraints
Master of Science (MS), Ohio University, 2013, Electrical Engineering (Engineering and Technology)
The purpose of this thesis is to track a vehicle using time difference of arrival (TDOA) measurements using two receivers and a road constraint. The nonlinear problem is solved using a direct solution, least squares, the extended Kalman filter and the Particle filter, and the results are compared. Also, different noise standard deviations have been considered so that the limitation of the filter with different noise can be studied. A generic problem has been simulated, in which the road distance, noise on the measurements and length of the road covered are a function of distance between the two TDOA sensors. The percentage of failed solutions was determined in a search grid to find the region in which the solution is feasible. The results show that both the least squares solution and the Kalman filter solutions experience large estimation errors, and that the solutions are often unstable. The Particle filter provides the best solution and does not diverge unexpectedly.

Committee:

Frank Van Graas (Advisor)

Subjects:

Engineering

Keywords:

Hyperbolic Measurements; Geometry Constraints; Particle filter

Ding, ErliSTUDY ON PARALLELIZING PARTICLE FILTERS WITH APPLICATIONS TO TOPIC MODELS
Doctor of Philosophy, Case Western Reserve University, 2016, EECS - System and Control Engineering
This thesis consists of studies in parallelizing particle filtering algorithms, various distributed computing frameworks and applications to information retrieval through topic models. We try to explore the possibility of a combination of these three seemingly unrelated areas in the thesis. The first part of the research investigates particle filtering theory and different parallelizing methods. This part proposes a novel resampling scheme for parallel implementation of particle filter. The proposed algorithm utilize a particle redistribution mechanism to completely eliminate the global collective operations, such as global weight summation or normalization. This algorithm achieves a fully distributed implementation of particle filters while keeping the estimation unbiased. The second part investigates the implementations of the particle filtering algorithms within two popular distributed computing frameworks, Hadoop MapReduce and Apache Spark. In addition to examining implementation, this part compares the pros and cons of the two different implementations and also discusses their respective usage. The third part considers the application of distributed particle filters to the area of information retrieval, in our case, topic modeling for batch and streaming documents. This part designs an auxiliary particle filter approach for learning and inference topics based on the dynamic topic model that captures the temporal structure of documents. In the experiment, we build an architecture for documents processing that includes both the batch processing power of MapReduce and streaming processing power of Spark. The input documents that are divided into time slices, document collections in each time slice share the same prior for their respective topic proportion and this prior is propagated over time. We use batch operations to preprocess and learn the models and then perform online inference streaming documents.

Committee:

Kenneth Loparo (Advisor)

Subjects:

Engineering

Keywords:

Particle Filter, Topic Model, Hadoop, Spark

Ahmadi, KavehDim Object Tracking in Cluttered Image Sequences
Doctor of Philosophy, University of Toledo, 2016, Engineering
This research is aimed at developing efficient dim object tracking techniques in cluttered image sequences. In this dissertation, a number of new techniques are presented for image enhancement, super resolution (SR), dim object tracking, and multi-sensor object tracking. Cluttered images are impaired by noise. To deal with a mixed Gaussian and impulse noise in the image, a novel sparse coding super resolution is developed. The sparse coding has recently become a widely used tool in signal and image processing. The sparse linear combination of elements from an appropriately chosen over-complete dictionary can represent many signal patches. The proposed SR is composed of a Genetic Algorithm (GA) search step to find the optimum match from low resolution dictionary. By using GA, the proposed SR is capable of efficiently up-sampling the low resolution images while preserving the image details. Dim object tracking in a heavy clutter environment is a theoretical and technological challenge in the field of image processing. For a small dim object, conventional tracking methods fail for the lack of geometrical information. Multiple Hypotheses Testing (MHT) is one of the generally accepted methods in target tracking systems. However, processing a tree structure with a significant number of branches in MHT has been a challenging issue. Tracking high-speed objects with traditional MHT requires some presumptions which limit the capabilities of these methods. In this dissertation, a hierarchal tracking system in two levels is presented to solve this problem. For each point in the lower-level, a Multi Objective Particle Swarm Optimization (MOPSO) technique is applied to a group of consecutive frames in order to reduce the number of branches in each tracking tree. Thus, an optimum track for each moving object is obtained in a group of frames. In the upper-level, an iterative process is used to connect the matching optimum tracks of the consecutive frames based on the spatial information and fitness values. Another problem of dim object tracking is background subtraction which is difficult due to noisy environment. This dissertation presents a novel algorithm for detecting and tracking small dim targets in Infrared (IR) image sequences with low Signal to Noise Ratio (SNR) based on the frequency and spatial domain information. Using a Dual-Tree Complex Wavelet Transform (DT-CWT), a Constant False Alarm Rate (CFAR) detector is applied in the frequency domain to find potential positions of objects in a frame. Following this step, a Support Vector Machine (SVM) classification is applied to accept or reject each potential point based on the spatial domain information of the frame. The combination of the frequency and spatial domain information demonstrates the high efficiency and accuracy of the proposed method which is supported by the experimental results. One of the important tools applied in this dissertation is Particle Filter (PF). The PF, a nonparametric implementation of the Bayes filter, is commonly used to estimate the state of a dynamic non-linear non-Gaussian system. Despite PF’s successful applications, it suffers from sample impoverishment in real world applications. Most of the recent PF based techniques try to improve the functionality of the PF through evolutionary algorithms in the cases of unexpected changes in the system states. However, they have not addressed the discontinuity of observation which is unpreventable in the real world. This dissertation incorporates a recently developed social-spider optimization technique into PF to overcome the drawback of previous methods in these cases. The problem of object tracking using multi-sensor data is a theoretical and technological challenge in the field of image processing which is presented as the final algorithm in this dissertation. Most of the conventional multi-sensor methods fail to track small dim objects in a cluttered background due to the lack of geometrical target information and unexpected large discontinuities in the measurement data. In this dissertation, a multi-sensor Swarm Intelligence Particle Filter (SIPF) is proposed in an environment covered by a set of multiple calibrated sensors with overlapping field of views. The proposed hierarchical method is divided into two levels. In the lower-level, SIPF is applied to locate the targets in each sensor based on the prior information. Each sensor reports the target position and its related fitness value to a dynamically selected central sensor. In the upper-level, the central sensor finds the best of the reported position for each target and broadcasts its position to all sensors at the lower level as the actual position of the target. Experimental results show this method is able to utilize multi-sensor data to produce highly accurate tracks in noisy datasets even in the case of large jumps or discontinuous observations well beyond the conventional tracking methods.

Committee:

Ezzatollah Salari (Committee Chair); Kim Junghwan (Committee Member); Jamali Mohsin (Committee Member); Carvalho Jackson (Committee Member); Eddie Yein Juin Chou (Committee Member)

Subjects:

Computer Engineering; Computer Science

Keywords:

Dim Object Tracking; Particle Filter; Particle Swarm Optimization; Multiple Object Tracking; Multi Sensor Object Tracking; Super Resolution

Chhatpar, Siddharth R.Localization for Robotic Assemblies with Position Uncertainty
Doctor of Philosophy, Case Western Reserve University, 2006, Mechanical Engineering
This dissertation deals with the class of robotic assemblies where position uncertainty far exceeds assembly clearance, and visual assistance is not available to resolve the uncertainty. Our research is motivated by actual assemblies from vehicular transmissions that fall under this class. For this class of assemblies, the focus shifts from the dynamics of the assembly to the problem of searching for part alignment. A novel idea is introduced to transform the search for part alignment into one of localizing the peg-hole misalignment on the hyper-surface formed in the peg-hole contact configuration space (C-space). This idea is developed into an intelligent localization strategy for resolving the uncertainty in the relative configuration of parts. The strategy is to explore the assembly contact C-space and match observations to a pre-acquired map of the C-space. The implementation of our localization strategy is described using both analytical and sampled maps of the contact C-space. Thus, one can either model the contact C-space using equations of the three-dimensional volumetric intersections of the mating parts, or sample it using a robot or CAD model. However, a sampled map does not provide a complete representation of the continuous contact C-space. Hence, the concepts of assembly sufficiency, goal region, and approximate localization are introduced to help in localizing sufficiently for assembly. With increasing dimensionality of the assembly uncertainty and small assembly clearances, the computational load becomes large and uneven over the localization period. An algorithm, termed the cell approach, is developed to implement the localization strategy in stages of increasing resolution, thus distributing the computational load more evenly. To make the localization strategy more robust, the application of particle filtering for robotic assemblies with position uncertainty was pioneered in this dissertation. Particle filtering is a probabilistic scheme that maintains a set of weighted particles, where each particle represents an estimate of the relative peg-hole configuration; it can handle errors in actuation and observation, and also errors in mapping. Moreover, the number of particles can be adjusted to accommodate the computational resources available. The ideas presented in this dissertation were validated with mathematical analyses, computer simulations, and actual robotic assemblies.

Committee:

Michael Branicky (Advisor)

Keywords:

Localization; Robotic Assembly; Particle Filter; Configuration Space; Sampled Maps

hart, charlesA Low-cost Omni-directional Visual Bearing Only Localization System
Master of Sciences, Case Western Reserve University, 2014, EECS - Computer and Information Sciences
RAMBLER Robot is designed to enable research on biologically inspired behavioral robot control algorithms. RAMBLER Robot tests the feasibility of autonomously localizing without typical sensors like wheel odometers or GPS. The primary objective is to independently, accurately, and robustly recover the path of a moving robotic system with only the lowest-cost sensors available off-the-shelf. Methods new and old are reviewed and tested on the real RAMBLER Robot hardware. The hardware and software necessary to use omni-directional camera measurements to decrease uncertainty regarding the position and heading of a small robot system are presented in detail. The RAMBLER Robot is shown to successfully localize within a small arena using three passive indistinguishable landmarks.

Committee:

Roger Quinn (Committee Chair); Francis Merat (Committee Member); Gregory Lee (Committee Member)

Subjects:

Computer Science; Robotics

Keywords:

omnicam; camera; omnidirectional; panoramic; catadioptric; spherical reflector; triangulation; power center; localization; particle filter; computer vision; raspberry pi; zumo; robot; robotics; low-cost; inexpensive; python; matlab; opencv;