Otto Fucik

Project of IPv6 Router with FPGA Hardware Accelerator

This paper deals with a hardware accelerator as a part of the Liberouter project which is focused on design and implementation of a PC based IPv6 router. Major part of the Liberouter project is the development of a hardware accelerator – the PCI board called COMBO6 and its FPGA design which allows processing most of the network traffic in hardware.

Particle rendering engine in DSP and FPGA

We present an algorithm for rendering 3D point-clouds, which exploits an FPGA chip coupled with a DSP processor on an experimental board. Point-clouds are sets of graphical data in 3D space, which seem to be more suitable for potentially many purposes than the most frequently, used triangle meshes. The actual experimental implementation, which verifies the concept and reports promising results, is also described.

A scalable cellular automata based microscopic traffic simulation

This paper presents a new model for simulations of very large scale traffic networks. The proposed model is based on microscopic cellular automata (CA) extended to eliminate unwanted properties of ordinary CA based models, such as stopping from maximum speed to zero in one time step. The accuracy of the model has been validated by comparisons with various fundamental diagrams. A parallel implementation developed using the proposed model allows for an almost linear speedup. This allows to run a simulation multiple in real-time, that the traffic state of very large scale networks can be precisely predicted, for example, with various scenarios.

Multiobjective Selection of Input Sensors for SVR Applied to Road Traffic Prediction

Modern traffic sensors can measure various road traffic variables such as the traffic flow and average speed. However, some measurements can lead to incorrect data which cannot further be used in subsequent processing tasks such as traffic prediction or intelligent control. In this paper, we propose a method selecting a subset of input sensors for a support vector regression (SVR) model which is used for traffic prediction. The method is based on a multimodal and multiobjective NSGA-II algorithm. The multiobjective approach allowed us to find a good trade-off between the prediction error and the number of sensors in real-world situations when many traffic data measurements are unavailable.

Evolutionary approach to calibration of cellular automaton based traffic simulation models

Microscopic traffic simulation models have become very popular in the evaluation of transportation engineering and planning practices in the past few decades. To achieve high fidelity and credibility of simulations, a model calibration and validation must be performed prior to deployment of the simulator. In this paper, we proposed an effective calibration method of the microscopic traffic simulation model. The model is based on the cellular automaton, which allows fast large-scale real-time simulation. For its calibration, we utilized a genetic algorithm which is able to optimize different parameters much better that a human expert. Furthermore, it is possible to readjust the model to given field data coming from standard surveillance technologies such as loop detectors in our case. We have shown that the precision of simulations can be increased by 20 % with respect to a manually tuned model.

Multiobjective selection of input sensors for travel times forecasting using support vector regression

In this paper we propose a new method for travel time prediction using a support vector regression model (SVR). The inputs of the method are data from license plate detection systems and traffic sensors such as induction loops or radars placed in the area. This method is mainly designed to be capable of dealing with missing values in the traffic data. It is able to create many different SVR models with different input variables. These models are dynamically switched according to which traffic variables are currently available. The proposed method was compared with a basic license plate based prediction approach. The results showed that the proposed method provides the prediction of better quality. Moreover, it is available for a longer period of time.

Fast and energy efficient AdaBoost classifier

The paper presents a new concept of creating an energy and computation effective AdaBoost classifier systems. The presented method is mainly novel in the way how it divides and accelerates an AdaBoost classifier into two parts -- a pre-processing and a post-processing unit. Pre-processing unit is designed to process a major part of the computational operations of the AdaBoost algorithm but it is also helps in an energy savings.

Calibration of Traffic Simulation Models Using Vehicle Travel Times

In this paper, we propose an effective calibration method of the cellular automaton based microscopic traffic simulation model. We have shown that by utilizing a genetic algorithm it is possible to optimize various model parameters much better than a human expert. Quality of the new model has been shown in task of travel time estimation. We increased precision by more than 25 % with regard to a manually tuned model. Moreover, we were able to calibrate some model parameters such as driver sensitivity that are extremely difficult to calibrate as relevant data can not be measured using standard monitoring technologies.

Automatic synthesis of small AdaBoost classifier in FPGA

Novel pre-processing units for AbaBoost classifiers are introduced which can improve performance and reduce power consumption in many image processing applications. An approach for automatic classifier synthesis to the FPGA is also described. The introduced classification architecture is intensively saving processing resources and very fast as well. Several optimization techniques that are used in the process of automatic synthesis are also shown.

Estimation of missing values in traffic density maps

The traffic density map (TDM) represents the density of road network traffic as the number of vehicles per a specific time interval. TDMs are used by traffic experts as a base documentation for planning a new infrastructure (long-term) or by drivers for showing a current traffic status (short-term). We propose two methods for estimation of missing density values in TDMs. In the first method, the problem is formulated relatively strictly in terms of quadratic programming (QP) and a QP solver is utilized to find a solution. The second, more general method is based on a multiobjective genetic algorithm which allows us to find a reasonable compromise among several objectives that a traffic expert may formulate. These two methods can work automatically or they can be used by a traffic expert for an iterative density estimation. Results of experimental evaluation based on real and randomly generated data are presented.