54 research outputs found
Robust statistics over Riemannian manifolds for computer vision
No full textThe nonlinear nature of many compute vision tasks involves analysis over curved nonlinear spaces embedded in higher dimensional Euclidean spaces. Such spaces are known as manifolds and can be studied using the theory of differential geometry. In this thesis we develop two algorithms which can be applied over manifolds.
The nonlinear mean shift algorithm is a generalization of the original mean shift, a popular feature space analysis method for vector spaces. Nonlinear mean shift can be applied to any Riemannian manifold and is provably convergent to the local maxima of an appropriate kernel density. This algorithm is used for motion segmentation with different motion models and for the filtering of complex image data.
The projection based M-estimator is a robust regression algorithm which does not require a user supplied estimate of the scale, the level of noise corrupting the inliers. We build on the connections between kernel density estimation and robust M-estimators and develop data driven rules for scale estimation. The method can be generalized to handle heteroscedastic data and subspace estimation. The results of using pbM for affine motion estimation, fundamental matrix estimation and multibody factorization are presented.
A new sensor fusion method which can handle heteroscedastic data and incomplete estimates of parameters is also discussed. The method is used to combine image based pose estimates with inertial sensors.Ph.D.Includes bibliographical references (p. 137-144)
How to make public works work : a review of the experiences
No full textThis paper reviews the experience with public works programs (PWPs) in several countries over the past 20 years to delineate use patterns and to determine the factors contributing to its use as a successful safety net program. The analysis shows that PWP have been used extensively in response to either a one-time large covariate shock, or repeated shocks. In low income countries, PWPs also have an antipoverty or poverty reduction objective. Our review shows that well designed and implemented PWPs can help mitigating income shocks; the program can also be used as an effective anti-poverty instrument. The paper examines the factors behind the observed wide variation in the effectiveness of the program in accomplishing its goals and identifies prerequisites for making PWPs successful safety net interventions capable of protecting the poor from income shocks, thus reducing both temporal and seasonal poverty, while creating useful public goods or services for the communities. For public works programs to be successful, it is important firstly to: a) have clear objectives; b) select projects that can create valuable public goods; and c) ensure predictable funding. Secondly, the success of the program depends critically on careful design and incorporation of all the key design features. Finally, a credible monitoring and evaluation system designed right upfront, prior to launching of theprogram can allow for mid course corrections and to respond to sudden changes which can inhibit effective implementation. The potential of the PWP program is enormous both in countries that have experiences with these programs and especially in countries that never used them. However, more research is needed investigation is needed to better understand the impact of PWPs, such as second round effects from the created assets, the impacts on the labor market, and their cost-effectiveness after factoring in both the immediate and second round benefits from its program.Safety Nets and Transfers,Rural Poverty Reduction,Labor Markets,Labor Policies,Public Sector Economics
Journal Self-Citation VIII: An IS Researcher in the Dual Worlds of Author-Reader and Author-Institution
No full textThis paper responds to the question of whether it is ethical for a journal editor to request an author to cite papers from a journal to which one is submitting an article. To craft a response to this question, two sets of relationships are explored. The first set is an author-reader relationship, and the second set is an author-institution or community relationship. In these dual relationships, the author is considered to be an IS researcher who publishes and disseminates knowledge through the channel of research journals. The reason for articulating these twofold relationships is to go beyond the common belief that the author is the sole and autonomous source of knowledge creation and distribution. We posit that: (1) an author cannot exist isolated from the reader, and (2) an author exists only as a part of an institutional system which opens and at the same time constrains an author’s knowledge production. In other words, an author is destined to create knowledge within the constrained system. For that very reason, it is important to understand the author as a function of conditional discourse of a specific institution. We conclude that editors’ requests for an author to cite papers from a journal to which one is submitting an article is ethically critical to: (1) build a good author-reader relationship, and (2) produce rich and plural knowledge which is “good” for advancing learning in the global community
Subspace estimation using projection based M-estimators over Grassmann manifolds
No full textAbstract. We propose a solution to the problem of robust subspace estimation using the projection based M-estimator. The new method handles more outliers than inliers, does not require a user defined scale of the noise affecting the inliers, handles noncentered data and nonorthogonal subspaces. Other robust methods like RANSAC, use an input for the scale, while methods for subspace segmentation, like GPCA, are not robust. Synthetic data and three real cases of multibody factorization show the superiority of our method, in spite of user independence.
Development of an ultra-wide band-based real-time vibrator tip locating system for intelligent concrete consolidation
No full textProper consolidation of concrete is critical to the long-term strength of concrete bridge structures. Vibration is a commonly used method to make concrete flow able and to remove the excessive entrapped air, therefore contributing to proper concrete consolidation. To introduce vibrations to freshly placed concrete, various tools such as internal vibrators are widely used in the construction industry. Producing a dense concrete without segregation with these tools requires an experienced vibrator operator. Inexperienced vibrator operators tend to over-consolidate or under-consolidate concrete. Many of these quality problems have their roots in the lack of quality control methods that can provide real-time feedback on the quality of concrete consolidation to vibrator operators. The proposed research involves the development of a real-time wireless sensing-based internal vibrator tracking system to support intelligent concrete consolidation operations. Specifically, the research team will explore the use of an Ultra Wideband (UWB) tracking system to realize precise localization of internal vibrators. Multiple tags will be attached to each vibrator for deriving its precise poses. Computer programs will be developed to track tags, to infer vibrator poses, and to visualize operators' vibration effort in real-time. Once a vibrator is tracked, the vibration location, time, and depth associated with this vibrator will be displayed on a computer in real-time. A vibrator operator can leverage such information to visualize the distribution of his vibration effort, and spot areas that may need mitigation actions. The new concrete consolidation tool will allow contractors to proactively address concrete consolidation issues, a problem common to many concrete construction projects.M.S.Includes bibliographical referencesby Raghav Krishnamoorth
Multisensory models for human spatial orientation including threshold effects
No full textThesis (S.M.)--Massachusetts Institute of Technology, Dept. of Architecture, 2010.This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections.Cataloged from student-submitted PDF version of thesis.Includes bibliographical references (p. 69-71).E-Observer, a stand-alone executable version of the Observer model developed by Newman and Oman (2009), was developed. The complicated structure of the Observer model and its parameters made this conversion challenging. The resulting Windows PC executable uses a publically available library (MATLAB component runtime v7.1 0). E-Observer parameters are limited to the preset choices in Observer. A hypothetical example of the use of E-Observer to analyze an aircraft accident radar trajectory data is discussed. Like many other dynamic models for human spatial orientation, Observer does not incorporate perception thresholds, which limits its use to relatively large stimuli and hence cannot be used for investigation of certain accidents and flight simulator design, which involve sub-threshold motions. The literature on motion thresholds is reviewed which suggests that vestibular perception thresholds are not mechanical thresholds, but are due to signal-in-noise phenomenon. As a fIrst step towards incorporating thresholds in Observer, modeling yaw perception thresholds was attempted and two detection models are proposed - a Matched Filter model and a Two-Threshold model. The Matched Filter detector model matches the noisy perception with a noise-free stimulus template and evaluates how much they correlate. Based on the correlation, the model fInally decides if the signal is present or not. However, this model applies only in cases where the subject is in an experiment, and knows the expected stimulus waveform. Grabherr et al (2008) proposed a high pass filter model for direction recognition thresholds based on their recognition data. This thesis explores an alternative modeling approach assuming that the CNS samples the angular velocity estimate and its derivative, and applies thresholds to both. Whether the motion stimulus is detected or not depends on how many of these samples cross the threshold level. The performance of both models was compared against the Grabherr et. al. data It was found that both models are able to approximate the 79.4% detection criterion for thresholds determined in Grabherr's study. However, the two threshold model does not assume that the subject knows the stimulus waveform. Supported by Project SA1302 by the National Space Biomedical Research Institute through NASA NCC 9-58.by Raghav Harini Venkatesan.S.M
Nonlinear Mean Shift for Robust Pose Estimation
No full textWe propose a new robust estimator for camera pose estimation based on a recently developed nonlinear mean shift algorithm. This allows us to treat pose estimation as a clustering problem in the presence of outliers. We compare our method to RANSAC, which is the standard robust estimator for computer vision problems. We also show that under fairly general assumptions our method is provably better than RANSAC. Synthetic and real examples to support our claims are provided. 1
- …
