Yasemin Yardimci Cetin

The Virtual Marathon: Parallel Computing Supports Crowd Simulations

To be realistic, an urban model must include appropriate numbers of pedestrians, vehicles, and other dynamic entities. Using a parallel computing architecture, researchers simulated a marathon with more than a million participants. To simulate participant behavior, they used fuzzy logic on a GPU to perform millions of inferences in real time.

A Cross-Cultural Analysis of Smartphone Adoption by Canadian and Turkish Organizations

The objective of this study is to identify the impact of cultural differences on adoption of smartphones in Canada and Turkey and investigate the differences in patterns between the adoption behaviors of the two countries. Sequential explanatory design mixed-method research strategy, which incorporates quantitative and qualitative approaches, was used in this research. A multi-group structural equation model analysis was conducted to assess the model based on the data collected from senior and middle managers at 213 and 141 private sector organizations in Turkey and Canada, respectively. Constant comparative method was used to analyze follow-up data that resulted from transcription of the interviews. Results show that national culture has a significant effect on adoption behavior and there are major differences in adoption characteristics between the two countries. For example, organizational characteristics, especially top management support, have a stronger effect on adoption of smartphones by organizations in Canada, while environmental characteristics, including competitive pressure, partner expectations, and customer expectations have a stronger effect on the adoption in Turkey. Implications of these results are discussed.

3D Scene Reconstruction Based on Robust Camera Motion Estimation and Space Sweeping for a Cultural Heritage Virtual Tour System

An efficient 3D reconstruction technique based on robust camera motion estimation and an improved version of the space-sweeping stereo reconstruction approach is presented. The proposed approach is focused on generation of usable and fully automatic reconstruction of wide-area scenes with the computational resources of a conventional PC. The aim is to use this technique to capture such scenes in 3D utilizing off-the-shelf equipment. Color information is finally added to the derived 3D model of the scene and the result can be converted to common 3D scene modeling formats. The 3D models are integrated with GIS technologies within a web-based virtual tour system.

Content based hyperspectral image retrieval using bag of endmembers image descriptors

This paper proposes a novel system for fast and accurate content based retrieval of hyperspectral images. The proposed system aims at retrieving hyperspectral images that have both similar spectral characteristics associated with specific materials and fractional abundances to the query image. It consists of two modules. The first module characterizes the query and the target hyperspectral images in the archive by two descriptors: 1) a binary spectral descriptor representing spectral characteristics of distinct materials 2) an abundance descriptor that contains the normalized cumulative fractional abundance information of the corresponding materials. Both descriptors are obtained by a novel bag of endmembers based strategy. The second module aims at retrieving hyperspectral images from the archive that are most similar to query image based on a hierarchical strategy which evaluates the spectral and abundance descriptors similarity. Experiments carried out on a benchmark archive of hyperspectral images demonstrated the effectiveness of the proposed system in terms of retrieval accuracy and time.

Signature based vegetation detection on hyperspectral images

In this study, the contribution of utilizing hyperspectral unmixing algorithms on signature based target detection algorithms is studied. Spectral Angle Mapper (SAM), Spectral Matched Filter (SMF) and Adaptive Cosine Estimator (ACE) algorithms are selected as target detection methods and the performance change related to the target spectral acquisition is evaluated. The spectral signature of the desired target, corn, is acquired from ASD hyperspectral library as well as from the hypespectral unmixing endmembers with a minimum angular distance to ASD signature. It is seen that the performance of the corn detection has increased significantly with the utilization of the closest endmember extracted from the hyperspectral data cube. Among all methods, SAM has been designated as the most successful method based on the Receiver Operating Characteristics (ROC) curves.

Shadow removal from VNIR hyperspectral remote sensing imagery with endmember signature analysis

This study aims to develop an effective regional shadow removal algorithm using rich spectral information existing in hyperspectral imagery. The proposed method benefits from spectral similarity of shadow and neighboring nonshadow pixels regardless of the intensity values. Although the shadow area has lower reflectance values due to inadequacy of incident light, it is expected that this area contains similar spectral characteristics with nonshadow area. Using this assumption, the endmembers in both shadowed and nonshadow area are extracted by Vertex Component Analysis (VCA). On the other hand, HySime algorithm overcomes estimating number of endmembers, which is one of the challenging parts in hyperspectral unmixing. Therefore, two sets of endmembers are extracted independently for both shadowed and nonshadow area. The proposed study aims at revealing the relation between these two endmember sets by comparing their pairwise similarities. Finally, reflectance values of shadowed pixels are re-calculated separately for each spectral band of hyperspectral image using this information.

Improvement of hyperspectral classification accuracy with limited training data using meanshift segmentation

In this study, the performance of hyperspectral classification algorithms with limited training data investigated. Support Vector Machines (SVM) with Gaussian kernel is used. Principle Component Analysis (PCA) is employed for preprocessing and meanshift segmentation is used to incorporate spatial information with spectral information to observe the effect spatial information. Pattern search algorithm is used to optimize meanshift segmentation parameters. The performance of the algorithm is demonstrated on high resolution Pavia University hyperspectral data.

Improvements on hyperspectral classification algorithms

This study investigates the effect of training set selection strategy on classification accuracy of hyperspectral images. This effect is analyzed in conjunction with three other factors, namely the use principal component analysis on the input data, and the use of spatial information and choice of classifier. Support Vector Machines (SVM) and Maximum Likelihood (ML) classifiers are used for demonstration. Meanshift segmentation and majority voting are used for inclusion of spatial information. The effect of the training data size and sampling strategy is demonstrated over the high resolution Pavia University hyperspectral data.

The effect of training data on hyperspectral classification algorithms

In this study, the performance of different hyperspectral classification algorithms with the same training set is investigated. In addition, the effect of the dimension and sampling strategy for the training set selection is demonstrated. Support Vector Machines (SVM), K- Nearest Neighbor (K-NN) and Maximum Likelihood (ML) methods are used. The contribution of using spatial information with spectral information is observed. Meanshift segmentation and window weighting methods are used for spatial information. High resolution Pavia University hyperspectral data and Indian Pines data are used in this study.

Out-the-window scene properties in pc-based helicopter simulators

PC based Flight Simulators (PC-FS) have been appeared as alternative training devices for pilot training to Enhanced Flight Simulators due to their low cost and absolute availability. Visuals presented in PC-FS are adequate for aircraft pilot training; this is not true for helicopter pilot training due to too low altitudes and speeds. Especially for hovering, increased visual quality is required because of extremely low altitude (3-15 feet) and small movements. In this project, two experiments were conducted by using simple PC-FS as a test platform and professional helicopter pilots as subjects in order to evaluate the effect of hyper texturing on hovering performance. Results have revealed that the level of texture resolution has no direct effect on hovering performance. Optimum texture resolution is dependent upon noticability, recognizability and size of the 2D objects presented by textured image.