Abstract:
In recent years, Traffic engineers widely use microscopic traffic simulation tools to evaluate traffic operational problems. It is due to the ability of traffic simulation tool to generate with different traffic problem in different scenario without even disrupting traffic condition on the road. Car following model is one of the most important driver behavior model that composes traffic simulation model. The road traffic situation of Bangladesh is heterogeneous in nature, where vehicles vary widely in static and dynamic characteristics and share the same width of carriageway. Therefore, the car following behavior under this condition is expected to be quite different from that of homogeneous or car dominated traffic. This justifies the development of car following models for heterogeneous road traffic condition under weak lane discipline.
However, in reality, the car following behavior of a driver is influenced by a vague perception of the stimuli and his decision making mechanism is driven by fuzzy rules gained by knowledge and experience. Therefore, the deterministic models fail to emulate the complex and multi ruled behavior of the driver in an uncertain environment of the road. Considering these issues, this study aims to develop a Neuro-Fuzzy model to depict car following behavior in heterogeneous road traffic condition.
In this study, car following data have been collected in the form of video footage of vehicular movements at one of the busiest roads of Dhaka city. From the video footage, car following parameters have been extracted by converting the two dimensional coordinates of vehicle trajectories of the video footage into real world positions or into pixel coordinates as required for different parameters along the road using Python programming language and OpenCV library. From these positions, various car following parameters such as distance, speed and acceleration were calculated. A car-car following model was developed by using the ‘Neuro-Fuzzy Designer’ toolbox of MATLAB. The model was calibrated through the Adaptive Neuro-Fuzzy Inference System of the same toolbox, in order to alter the membership functions such that the models are able to replicate the real scenario of vehicular movement accurately.
In order to validate the models, at first few criteria were defined to perform statistical analysis of the proposed model. Furthermore, two new models were introduced and statistical analysis was done for the new models. At last, performance of the proposed model was observed by comparing the errors of all the models.
