Selected Abstracts
_______________________________________
VBR Video Traffic Modeling and Synthetic Data
Generation Using GA-Optimized Volterra Filters
Adel Abdennour
Electrical Engineering Dept, King Saud University
P.O. Box 800, Riyadh 11421, Saudi Arabia
Email: adnnour@ksu.edu.sa,
ABSTRACT:
Variable Bit Rate (VBR) video traffic models, which accurately represent the traffic characteristics and statistical properties of real videos, can provide significant knowledge about expected traffic behavior. This knowledge can be used in the development of effective control schemes and improved network quality of service. An interesting class of models based on the idea of generating a number of chi-square sequences, by passing a Gaussian AR process through a simple nonlinearity, was recently introduced. The gamma process in this class of models is obtained by a linear combination of the chi-square sequences. This model is simple and allows for arbitrary selection of both the auto-regressive (AR) model and the shape parameter of the gamma probability density function (pdf). However, the AR filter order is chosen mainly on a trial-and-error process. In addition, while the approach uses a linear combination of K chi-square sequences, it fixes (K-2) coefficients and solves for only the remaining two because it has more equations than unknowns. Occasionally, the resulting solution is not feasible and additional trials for different solutions are required. It is, therefore, the objective of this paper to use genetic algorithms (GA) to provide a more systematic approach to find the various model parameters. The paper also presents a thorough statistical analysis of the generated synthetic data in order to qualify its suitability for representing MPEG video traffic. A comparison with published results is carried out in terms of how close are the means, standard deviations, and the autocorrelation functions to those of the real data. The comparison over 10,000 replications and for a number of different video traces reveals that a significant improvement can be achieved in almost all measures and for almost all the movies tested.
Artificial Neural Networks Application to the Estimation
of Vehicle Headways in Freeway Sections
Adel Abdennour and Ali S. Al-Ghamdi
College of Engineering, King Saud University
P.O. Box 800, Riyadh, 11421, Saudi Arabia
Abstract
Vehicle headways play a role of paramount importance in many traffic engineering applications. They provide operators of transportation systems with information for selecting and designing traffic control strategies and safety measures. Their role will undoubtedly even increase particularly in the intelligent transportation systems. Modelling vehicle headways, as a probability distribution function or time series, has been the focus of a big number of research projects, most of which was dealing with the statistical approach. This paper presents an Artificial Neural Networks (ANN) alternative to the classical techniques. Two networks were designed: one for the time series problem and the other is for the general probability distribution function. Simulation of the two networks with data gathered from nine different freeways in Riyadh revealed that accurate models can be achieved. The network was trained with all the data mixed up. However, it was able to reproduce the behavior of any single freeway.
Key words: Neural Networks, freeways, vehicle headway, general probability distribution
A Long Horizon Neuro-Fuzzy Predictor
For MPEG Video Traffic
Adel Abdennour
College of Engineering , King Saud University
P.O. Box 800, Riyadh 11421, Saudi Arabia
E-mail: adnnour@ksu.edu.sa
Abstract
This paper investigates the long-term prediction of MPEG video traffic. Predicting such traffic over a long horizon is important for today’s fast networks and internet multimedia services. In comparison with short-term prediction, long-term prediction of video traffic is yet to be explored especially for MPEG-4 coded videos despite its effectiveness in a number of important network-edge applications such as dynamic bandwidth allocation, quality of service (QoS) control, and network management and planning. The main reason for the shortage of publications in such area is the difficulty of the problem especially when classical or widely used prediction techniques are the ones to be employed. Prediction results in this paper are obtained using a simple neuro-fuzzy system and are compared to the classical normalized Least Mean Squares (LMS) technique. The neuro-fuzzy predictor is capable of predicting various real MPEG-4 real-world video traffic, for up to hundreds of frames in advance, with promising accuracy.
Short-Term MPEG-4 Video Traffic Prediction Using ANFIS
Adel Abdennour
College of Engineering, King Saud University
P.O. Box 800, Riyadh 11421, Saudi Arabia
Phone: (966) 1-467-8594 Fax: (966) 1-467-6757
adnnour@ksu.edu.sa,
Abstract
Multimedia traffic and particularly MPEG coded video streams are growing to be a major traffic component in high speed networks. Accurate prediction of such traffic enhances the reliable operation and the Quality of Service (QoS) of these networks through a more effective bandwidth allocation and better control strategies. However, MPEG video traffic is characterized by a periodic correlation structure, a highly complex bit rate distribution, and very noisy streams. Therefore, it is considered an intractable problem. This paper presents a neuro-fuzzy short term predictor for MPEG-4 coded videos. The predictor is based on the Adaptive Network Fuzzy Inference System (ANFIS) to perform single step predictions for the I, P, and B frames. In addition, short-term predictions (one, two, and four steps ahead) are also examined using smoothed signals of the video sequences. The smoothed time series of the coded videos are obtained using a moving average. The ANFIS prediction results are evaluated using long entertainment and broadcast video sequences and compared to those obtained using a linear predictor. The results are also compared to those obtained using a neural network approach published in the literature. ANFIS is capable of providing accurate prediction and has the added advantage of being simple to design and to implement.
Evaluation of Neural Network Architectures
For MPEG-4 Video Traffic Prediction
Adel Abdennour
Department of Electrical Engineering
King Saud University
P.O. Box 800, Riyadh 11421 Saudi Arabia
E-mail: adnnour@ksu.edu.sa,
Abstract
Multimedia applications and particularly MPEG-coded video streams are becoming major traffic components in high speed networks. Traffic prediction is important in enhancing the reliable operation over these networks. However, MPEG video traffic exhibits periodic correlation structure and a complex bit rate distribution, making prediction difficult. Neural networks can effectively be used to overcome such problem. In the literature, the problem has been mostly evaluated using standard feed-forward neural networks. However, a significant improvement can be expected using different types of neural networks. In this paper, six separate neural network predictors (including feed-forward) that can predict the basic frame types of MPEG-4: I, P, and B are developed and evaluated using long entertainment and broadcast video sequences. The performance is also compared to the widely used linear predictor. Comparison with results published in a recent work is also presented.
A NEURO-FUZZY MODELLING OF GAS HOLDUP IN BUBBLE COLUMNS OPERATING IN THE HOMOGENEOUS FLOW REGIME
Adel A. Abdennour1 and Waheed A. Al-Masry2
1Department of Electrical Engineering, King Saud University, Riyadh, Saudi Arabia
2Department of Chemical Engineering, King Saud University, Riyadh, Saudi Arabia
Abstract
Gas holdup measurements were made in a 15 cm bubble column with air-water system. The passive acoustic waves associated with gas bubbling phenomena were measured using hydrophone and modelled using an Adaptive Neuro-Fuzzy Inference Systems (ANFIS). Using the ANFIS model, gas holdup was successfully predicted with errors less than 1.5%. In addition to its accurate estimation results, the model uses only two statistical properties of the passive acoustic waveform (i.e., median and standard deviation), in addition to the sensor position, rather than the whole lengthy signal. The resulting architecture is a three-input-one-output network, simple enough for practical applications. The results reveal that ANFIS can be an easy and effective technique in modelling gas holdup in bubble columns with moderate operating conditions such as bioreactors.
Keywords: Bubble column, acoustic, hydrophone, ANFIS, gas holdup.
Real-Time Vehicles Detection and Traffic Parameter
Extraction for Highway Surveillance
S. Al-Garni, and A. Abdennour
King Saud University, College of Engineering,
Electrical Engineering Department, P. O. Box 800, Riyadh 11421
E-mail: algarnis2@yahoo.com, adnnour@ksu.edu.sa
Abstract
Real-time freeway monitoring is becoming more and more important due to the steadily increasing traffic and the limited capacity of available highway infrastructure. Proper monitoring of highway traffic can improve the performance of such infrastructure. Highway monitoring is a challenging practical problem and a number of different approaches have been proposed in this active research area. This paper proposes a real-time vision-based monitoring system. The proposed system takes advantage of the powerful artificial neural networks (ANN) for vehicles detection and counting. The detection process uses the freeway real time images and starts by automatically extracting the image background from the successive frames. Once the background is identified, only an update is required for subsequent processing to accommodate expected environmental and light changes. The system is implemented and tested on the busiest freeway in Riyadh (King Fahd Road) and achieved a high correct detection rate in the order of 98%.