Kernel-based modeling and optimization for density estimation in transportation systems using floating car data

Traffic congestion is one of major problems in numerous cities especially in urban areas. An appropriate solution comes from the modeling of traffic data and understanding the congestion characteristics. Various methods were developed to solve this problem, however, still necessary to develop new approaches. In this paper, a kernel-based density estimation method is utilized to extract the congestion spots in urban areas based on collected position samples with time-stamp from floating car data. A probabilistic framework is developed to model the traffic data with generalized Gaussian density and then to find optimized weights of kernels in an approximation function, centered at points-of-interest by minimizing the Cramer-von Mises distance between localized cumulative distributions of mixture of Dirac distributions of position samples and Gaussian mixtures of points-of-interest in a pre-defined time window. The approximation density function by optimized kernels' weights can be used to estimate the mobile vehicles density in a specific time and space. Modeling the traffic data to extract the required parameters improves the performance significantly. The proposed method is applied to real measurements and can be implemented in real time in traffic management systems.