Abstract: In video-based vehicle monitoring systems, image noises seriously influenced the results of vehicle recognition, background learning, and so on. The existing filtering algorithms could not meet both the speed and filtering effect requirements of these systems. After comparing the effect and speed of vehicle recognition without filtering and with median filtering or mean filtering, the mean value filtering was selected for further study. By appropriately reducing the amount of calculation, the speed of the new filtering algorithm was increased through improving the processes of data accessing and calculating, and the algorithm was realized with Visual C++ 6.0. Experiment results show that the speed of the new algorithm is 55% faster than box filtering.