Vision-based automatic vehicle counting system using motion estimation with Taylor series approximation

Automatic tracking vehicle in urban traffic video surveillance is a challenging problem in computer vision. Although many issues have been solved, some are still unsolved, such as video surveillance problem of complex traffic intersection in congested condition. In this paper, we develop a vehicle counting system using a motion estimation with Taylor series approximation with embedded virtual entering and exiting boxes. The result shows that the system provides the counting success rate as high as 100% and the lowest counting rate is 14.29%. The mistakes are from the wrong direction prediction because of the very complex traffic condition.