T. Saar

Adaptively undersampled, circular histogram based image processing for rotation invariant coin detection

As in many real time signal processing application in image processing the computation time and the number of operations is very critical. Correlation as the most known and most widely used algorithm in image matching has a tendency to grow the number of operations rapidly when the range of discretion increases. In this paper adaptively undersampled circular histogram based image processing algorithm for the validation of fast moving and rotating coins has been developed and evaluated. The combination of adaptive undersampling and circular area histogram matching makes the method highly accurate and rotation invariant while significantly reducing the calculations operations needed for the validation of fast moving and rotating coins. This method has been tested on real euro coins and the results show that this method is suitable for accurate rotation invariant coin verification system.

Chirp-based impedance spectroscopy of piezo-sensors

Impedance spectroscopy is widely used in various test& measurement fields and applications. In current paper a novel approach of excitation with chirp signal with simple time-domain analysis of response signal for such task is introduced and described with examples of measurement of the electro-mechanical impedance of piezo-sensors, in the frequency range of several hundred kiloherz. In one approach smoothed separately (by Savitzky-Golay filter) of excitation voltage and (current) response signals in a sliding window are found and their ratio is used for impedance module estimation, at corresponding time and frequency values. Alternative method for vector measurement of impedance spectra in time domain has also tested.

Adaptively undersampled image processing for fast multiline laser detection

In many image processing applications one of the main problem is how to create algorithms, which are fast enough to be implemented on embedded (with limited resources) computing platforms, like DM3730 processor based development boards (Beagleboard-XM). One such image processing area is related to laser scanners, which are usually implemented on multi-core high performance personal computers. In this paper an adaptively undersampled image processing method for fast multiline laser detection has been proposed. The combination of adaptive undersampling and multiline laser detection significantly reduces the calculation operations, needed for laser line detections, thus making this method highly suitable for limited embedded platforms. This method has been tested with real multiline laser images on pavement surface, on 4 different computing platforms and the results are presented at the end of the paper.

Simple DSP interface for impedance spectroscopy of piezo-sensors

An interface for impedance spectroscopy measurements of piezo-sensors has been developed, for a digital signal processor of Delfino series of Texas Instruments. DAC of the interface is based on a 16-bit PWM with extra 8-bit part of “high resolution” (of picoseconds), with external simple (3-rd order) analog filter. Internal 12-bit ADC is converting at rates of up to 10 MS/s with 2 simultaneous sample-and-holds. So, with few extra components, high-performance analog interface has been “improvised” (developed and investigated) with a frequency range 10 kHz −400 kHz about 0,1% of the full scale resolution and repeatability of measurements.

Methods for increased sensitivity and scope in automatic segmentation and detection of lung nodules in CT images

We propose two methods for lung segmentation and nodule detection in the initial stage. The lung segmentation method is based on a combination of masks, flood fill algorithm, and morphological closing operation. The nodule detection method is based on a multi-level thresholding process combined with various feature extraction techniques. The methods are evaluated with the new Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) dataset, which increases the number of tested cases, and thus the confidence, reliability and robustness of the results as compared to previously reported research. We aimed at detecting all nodule challenges present in the LIDC-IDRI (nodules<;3 mm and nodules with 3-30 mm diametric size). Statistically, we have detected 3058 nodules in the initial stage (including solid, non-or part-solid, juxta-vascular, juxta-plural, and well-circumscribed nodules). To the best of our knowledge and given the results presented in other previous works (for the final stage), our results indicate that we can detect more nodules in the initial stage than ever reported, and that we can handle all types of nodule challenges in LIDC-IDRI. Specifically, we have achieved a sensitivity of 85% on the complete LIDC-IDRI dataset. On the smaller dataset of 60 CT scans with 315 nodules, we detected 301 nodules and achieved a sensitivity of 95.5%.

A tool for lung nodules analysis based on segmentation and morphological operation

Lung cancer has become the foremost cause of high mortality rate in the western world. Subsequently, many image databases are being developed to enhance the investigation of nodules detection techniques in the early stages of lung cancer. Notably, Lung Image Database Consortium (LIDC) is widely recognized as one of the truthful databases but the efficient assessment of nodules on LIDC based Computed Tomography (CT) slices remains challenging. A novel, simple to use and efficient computer based application is proposed for the visualization, marking and segmentation of nodules on lung CT images. Greyscale thresholding and morphological operation (erosion) are applied to achieve adequate nodules segmentation on CT images. Furthermore, an extension enabling 3D visualization is also introduced. This application provides a solution to the challenges related to the segmentation and analysis of nodules in the CT images.

Experimental modal analysis of maritime composite panel

The investigation of dynamics of the real-life natural vibrations in structures, compared with the simulated by finite-element model frequency values and their variations, could be used for monitoring of the structural health of the composite materials. Experimental modal analysis of an example maritime composite panel is presented in the current work. The previously developed setup of the measurement system of the acceleration data for the structural health monitoring has been re-used. The results are analysed and compared with theoretically simulated results. The appropriate settings of modal analysis and their initial values for modal analysis techniques have been identified experimentally, to extract (estimate) the natural frequencies with reasonable accuracy and computational complexity.

Extraction of the variable width laser line

The aim of this paper is to develop a laser line detection algorithm which reduces the uncertainties what are caused by the unknown laser line width on the laser scanner images. Due to the fact that different surfaces reflect light in different ways the laser line on captured images is not always with constant width. The line width is also depending of the laser optics used. In 3D laser line scanners a correct detection of laser line is essential, because the laser line extraction accuracy affects greatly the precision of the developed 3D laser scanner. In this paper a method to use different standard deviations for inverted second order Gaussian derivative convolution mask to find the optimal standard deviation for every line section has been proposed. The results show a great dependency between standard deviation of the convolution kernel and laser line width. The experiments show that the optimal standard deviation of the convolution kernel is not constant over the whole laser line. Furthermore the optimal standard deviation for every line section reduces the uncertainties that were caused by the unknown laser line width.