Yongkwon Kim

Building Recognition for Augmented Reality Based Navigation System

In this paper, we propose a building recognition algorithm for the next generation navigation system which is to use captured video from the camera equipped in moving vehicle. After masking out the road area and vehicles on the road, we used edge segments of small size blocks within an image frame to remove background objects, trees, etc. Search range determination and edge tracing technique were also developed to determine the building area as the final step. The experiments show promising results such as 88.9 % detection rate on the average and 95.9% for navigation video without trees and signs. The proposed algorithm can be also applied to point-of-interest (POI) detection for various telematics applications.

Optimal Optical Conditions and Positioning Scheme for an Ultrahigh-Resolution Silicon Drift Detector-Based Gamma Camera

In this study, we optimized the optical conditions and associated positioning scheme for an ultrahighspatial-resolution, solid-state gamma detector. The detector module consisted of an array of seven hexagonal silicon drift detectors (SDDs) packed hexagonally and coupled to a single slab of crystal via a light guide glass. Because the optical behavior and requirements of the detector module and noise characteristics of the SDD sensor are different from those of conventional photomultiplier tube (PMT)-based detectors, the following parameters were studied to determine the optimum condition: scintillator selection, the effect of cooling on signal-to-noise ratio (SNR), the depth dependence of the scintillation light distribution, and optimum shaping time. To that end, a modified, Anger-style positioning algorithm with a denoise scheme was also developed to address the estimation bias (pincushion distortion) caused by the excessively confined light distribution and the leakage current induced by the SDD sensor. The results of this study proved that the positioning algorithm, together with the optimized optical configuration of the detector module, improves the positioning accuracy of the prototype detector. Our results confirmed the ability of the prototype to achieve a spatial resolution of about 0.7mm in full width at half maximum (FWHM) for 122 keV gamma rays under the equivalent noise count (ENC) of 100 (e- rms) per SDD channel. The results also confirmed NaI(Tl) to be a more desirable scintillator for our prototype with an energy resolution performance of about 8%.

Traffic Light Detection Using Rotated Principal Component Analysis for Video-Based Car Navigation System

This letter presents a novel approach for traffic light detection in a video frame captured by an in-vehicle camera. The algorithm consists of rotated principal component analysis (RPCA), modified amplitude thresholding with respect to the histograms of the PC planes and final filtering with a neural network. The proposed algorithm achieves an average detection rate of 96% and is very robust to variations in the image quality.

Building detection in augmented reality based navigation system

In this paper, we propose a building detection algorithm for video based navigation system which overlays driving directions on the captured video from in-vehicle camera. We used edge segments of small size blocks within an image frame to remove background objects, trees, etc. Search range determination and edge tracing technique were also developed to determine the building area as the final step. The experiments show 88.9 % detection rate on the average and 95.9% for navigation video without trees and signs. The proposed algorithm can be also applied to point-of-interest (POI) detection for location based services (LBS) in ubiquitous computing environment.

A peak detection in noisy spectrum using principal component analysis

A spectrum of a radio isotope (RI) contains a single or multiple photo-peaks and radio-activities of all energy levels. These characteristics of each RI source are measured by radiation monitor (RM) systems. However, if the radiation count is extremely low and source to detector distance is too far, we cannot acquire good spectroscopic results for RI identification by RM devices while we still able to measure some counting statistics. Thus, precise peak detection in noisy spectrums is one of the most important tasks in the RM system. In this study, we developed an accurate peak detection method based on wavelet decomposition followed by principal component analysis. We used a discrete wavelet transform (DWT) for reduction of unnecessary high frequency noises in low counts spectrums. To reduce effect of a background radiation, we made a background template using a pre-measured background spectrum and calculated square errors for suppressing a background of low energy levels and maintaining true photo-peaks. Finally, we analyzed pre-processed data and detected photo-peaks using PCA. We measured Cesium 137(Cs-137) and Barium 133(Ba133) with 1 and 10 micro curies collected from the various distance. Each spectrum was collected for a second and total 60 sets were stored for each isotope. Results of our research show that the proposed algorithm achieves high sensitivity and specificity, proving that the algorithm is appropriate for RM systems.

Registered collimation for pixellated SPECT detectors

We investigated a high sensitive collimator (registered collimator) design and associated image reconstruction methods. Its opening was designed to align with the CZT pixel by having same pitch and shape. By the combination of wide opening holes and low septa height, the sensitivity of the collimator can increase dramatically. However, it may also lead to loss of system resolution. The goal of this study is 1) to determine the optimal parameters of the collimator which can increase sensitivity and decrease the charge sharing effect, and 2) to develop system matrix for iterative image reconstruction that can adequately model the proposed collimation scheme to achieve high resolution recovery in spite of using wider collimation holes. Thickness and height of septa in the registered collimator were analytically determined to 1 mm which blocks 97% of 140keV gamma rays and 5.48 mm which indicates 0.2 % efficiency that equivalently means about 10 times higher counts than low energy high resolution(LEHR) collimator. Then 2D fan beam-based geometrical response function(GRF) was developed to model the wide opening holes in the registered collimator. The GRF was included in the system matrix of ordered subset expectation maximization(OSEM) method to compensate the loss of resolution. For performance comparison, GATE simulation was performed using a 5 mm diameter hot sphere phantom, which is located at 1 cm off-centered radial position, and a 40.36 by 40.36 mm CZT detector module. The preliminary results showed that the proposed collimator achieved about 13 times more counts and improved image resolution. In the future, we will keep our experiments with3D cone beam model as well as 2D fan beam model using more realistic phantoms.

Image registration for PET/CT and CT images with particle swarm optimization

The objective of this research is to develop 2D/3D registration algorithm for PET/CT and CT or MR images acquired by different systems at different times. We matched two anatomical images first (one from PET/CT and the other from standalone CT or MR) that contains affluent anatomical information. Then we geometrically transformed PET image according to the result of transformation parameters calculated by previous step. We developed two stages of registration algorithm. The first stage is global registration. It is consists of 4 independent steps. After selection of reference and target images different data types and ROI of images have been normalized in the preprocessing step. As a next step, target image is geometrically transformed then the similarity between two images has been measured quantitatively. The optimization step updates transformation parameters to find the best matched parameter set efficiently. In second stage that is called fine adjustment, we introduce feature based registration algorithm. The features of each image slices are extracted by independent component analysis(ICA) and the extracted feature plane is used to measure the similarity. B-spline based freeform deformation is done to form a registered image as a final step. The result of proposed algorithm shows good agreement of images that between PET/CT to CT and PET/CT to MR. We will expand the application of the algorithm to different imaging modalities.

Extended TCAM for Adopting Range and Negation Rules in Packet Classification

General packet filtering systems using TCAM have some limitations such as inefficient searching the range and negation rules. To improve these problems, we propose the extended TCAM architecture and the efficient scheme for converting the rules into smaller TCAM entries. The proposed TCAM architecture has additional RC blocks for negation rules and Gray code generation block for range rules. The converting algorithm takes advantage of the Gray codes symmetric characteristics to reduce the number of TCAM entries for range rules and utilize the functionality of RC blocks for negation rules. Functional tests and performance analysis of the proposed TCAM are conducted by using a FPGA design tool and the simulation results with the SNORT rules show that the proposed TCAM architecture and algorithm reduce 93% of TCAM entries at the expense of little hardware overhead of 2 bits per a memory word

Investigation pseudo 3D imaging method for small field of view (SFOV) using a SFOV gamma camera

We have investigated a limited angle gamma imaging system that enables pseudo 3D reconstruction of small objects such as thyroid, prostate and etc. In this particular study, we are interested preliminary results using small phantom imaging that provides 3D imaging information by typical planar imaging configuration. To achieve the goal, we developed a system equipped with a multi-pinhole collimator and the corresponding reconstruction method that provides transverse slices of the small field of view (SFOV). Our group has been developing a high resolution SFOV gamma camera system. Simulation studies have been conducted using a geant4 application for tomographic emission (GATE) simulation tool. We make three sphere phantoms which lay through the center of FOV with different depths. An eight-pinhole collimator with 2mm diameter aperture was configured for the simulation. For image reconstruction, system matrix for a maximum likelihood expectation maximization (MLEM) was developed based on the geometry of the collimator. The projector and back-projector were separately implemented based on the ray-driven and voxel-driven methods respectively. Simulations have been conducted with respect to 60, 90, 120 degree angle spans and 3, 6, 9 degree angle pitches. Preliminary results showed that the system distinguished the three sphere phantoms (2.5 cm depth). In addition, Contrast enhancement of phantom regions was observed as number of iteration increased. In the future, we will continue to investigate the optimum pinhole configurations pinhole size, angle span, and angle pitch, etc.

Data acquisition system for multi channel silicon drift diode-based detector

We have developed front-end electronics and data acquisition unit for a silicon drift diode (SDD) based prototype detector. The prototype detector consists of an array of 67 SDDs. The array is bump-bonded on a mother board which is cooled by Peltier-water based cooling system. In this study, we describe the function of front-end electronics (including preamplifier, shaper and peak stretcher units) and a data acquisition (DAQ) unit. The hardware consists of three major blocks: 1) a preamplifier board that amplifies the SDD signal, 2) a shaping and stretcher board that shapes event signal and stretches it, and 3) DAQ board that converts analog signal to digital and acquires digitized data. The preamp board amplifies 67 channel SDD signals with OP-amps and controls the jFET of each SDD to optimize reset interval and gain. The shaping board consists of 5 sub-blocks that are shaper, peak stretcher, MUX, discriminator and SHP_ctrl. It shapes the preamplifier signal and generates event signals. The DAQ consists of an array of A/D converters and FPGAs. In our FPGA simulation, the system uses 62% of memory, 21% of LOGIC slice. We evaluated and verified 3 boards with including FPGA program. In the result we found and settled some problems which cross talk and noise. However, we meet two problems that are SDD sensor board and synchronous reset circuit. We will fix some problems and evaluate the detector performance such as spatial resolution and energy spectrum, etc. We will also include the results of phantom study later.