Gordana Velikic

Musical note segmentation employing combined time and frequency analyses

In this paper we describe a musical stream note segmentation method that employs time-domain and frequency-domain analysis methods working in conjunction. The method has two demonstrated benefits: first, it leads to reliable results with very low probability of either missing or of falsely detecting notes, and second, it has temporal resolution on the order of 1 msec. We have applied the method to a variety of monophonic musical instrument recordings, including the clarinet, piano and violin, with results that vary from 95% to 100% accuracy.

Methods to extract respiration information from ECG signals

The reliability of respiratory signal extraction methods from electrocardiogram (ECG) data is investigated. To provide a reference, a breathing sensor using a piezo-electric cable was designed and constructed. Two previously established methods [1], QRS amplitude modulation and respiratory sinus arrhythmia (RSA), are adapted to include the PQRST complex of the ECG. In addition, a third method, which we call the ECG mean, is proposed. The results show that interval methods perform better than envelope methods. However, the ECG mean method performs similarly to the interval methods.

Combined testing approach: Increased efficiency of black box testing

This paper presents a methodology that combines both - manual and automatic black box testing. Advantage of the hybrid approach is that combined benefits of each methodology mitigate the weaknesses of a single methodology. This increases the efficiency of black box testing. The presented methodology generates test plans for new functional testing campaign which is based on the test history results, and thus the test times are shorter with same testing quality.

The multimedia devices portable automated test system

Rapid development of new technologies adds to an architectural complexity of multimedia devices. The last generations of multimedia devices combine features of multiple devices into a single device. Complex devices require more testing and verification methods, and more time to find a fault. Very often inoperative device has to be transported to operators facilities for complete diagnostics. Transportation of inoperative devices to repair/service centers prolongs time from a problem reporting to problem correction, increases costs, and adds to the customers discontent. Field test system comprises various input interfaces which cover multimedia devices audio/video output, and allows functional verification, wide A/V analysis and picture quality measurement. Field test system is ideal for verification of STB/DVD/Blu-ray players devices at customers homes or retailers offices and also for monitoring of head-end signal quality in broadcast systems.

Predicting the near-future impact of daily activities on heart rate for at-risk populations

In this paper we demonstrate the ability to predict changes to heart rate due to changes in levels of activity, up to an hour into the future. Activity levels are calculated from data collected by a worn accelerometer for a person performing daily activities outside a laboratory environment. People with congestive heart failure must take care not to excessively stress their heart. This can be a challenge due to the difficulty of predicting how much stress an activity is exerting on the heart. We propose to model the relationship between motion and heart rate and thus to enable the prediction of heart rate changes prior to performing an activity. We explored three methods to predict current and future heart rate from activity level: a continuous state Kalman Filter, two simple linear models, and a nonlinear model given in the literature [5]. The results from healthy subjects and subjects with congestive heart failure show that using the proposed models, the heart rate can be predicted an hour into the future using accelerometer data.

Blind DIBR-synthesized image quality assessment based on sparsity features in morphological multiscale domain

In this paper a no-reference image quality assessment (IQA) metric for DIBR-synthesized images is proposed. Sparsity based features of morphologically decomposed image subbands are used to estimate distortion level in images. A General regression neural network is utilized to calculate quality score. The performance is evaluated using publicly available IRCCyN/IVC DIBR image database. Experimental results show that proposed metric accords with human subjective judgment.

Algorithm optimization framework for Advanced Driver Assistance Systems

With the rapid technology development of systems that assist drivers, the lack of uniform support for efficient algorithm development has emerged. We propose a universal Software Development Kit (SDK) concept with common optimization routines that can be applied to typical Advanced Driver Assistance Systems (ADAS) algorithms. The evaluation confirmed that the SDK decreased execution time for ADAS algorithms and works seamlessly across different embedded architectures.

A solution to software scalability across ADAS platforms

Recently Advanced Driver Assistance Systems (ADAS) use cases are growing in number and ubiquity with no common approach to software design and implementation to facilitate Advanced Driver Assistance Systems development. Thus we are presenting our scalable software model where we focus on data acquisition, processing units, and action intelligence. In this software module we present our Advanced Driver Assistance Systems Middleware design and communication with Advanced Driver Assistance Systems modules. The middleware design supports acquiring and processing data on different units with separation of concerns between the safety critical and non-critical data. This software model enables development of complex Advanced Driver Assistance Systems modules with higher efficiency based on standardized data types and data exchange between modules. The goal of this model is to help a researcher to reuse modules on different Advanced Driver Assistance Systems platforms.

Lua-based self-management framework for Internet of Things

This paper presents a solution for a self-management system framework. Implemented solution enables network administrators to create, edit and execute Lua configuration scripts. Key benefits are: Evaluation of Luas deployment in autonomic systems and definition of the event-model used for the configuration application.

Employing the mutual phase coherence of the overtones in musical sounds for audio source separation and reconstruction

Source separation from a monaural mixture of musical sounds is challenging. When only a single mixture of multiple sources is available, the method of independent component analysis may not be straightforwardly applied, and proposed source separation methods have relied on previous knowledge of the independent sources. Our method employs mutual correlations of the overtone phases of musical sounds to ascribe spectral features to individual sources. The phase versus time is computed from the analytical signal representation of the output of a set of adaptive band‐pass filters that perform initial spectral separation. A small number of cleanly separable partials, usually the fundamentals, are then employed as references to which the remaining filter‐bank outputs are correlated. To separate overlapping spectral features and ascribe the proper amount of power to each source, the phase of the reference signal is scaled to the harmonic frequency, and the amplitude of each overtone is estimated by computing the ...