Dynamic traffic simulation tools are more and more used to help traffic managers and urban planners to make decisions. Therefore, simulation tools users now ask to research a validated methodology guaranteeing that simulation results are trustable. This paper contributes to identify and correct one of the possible lacks of detailed calibration and validation process of car-following models, the data errors of individual trajectory data. Some studies addressed the problem of filtering trajectory data. We propose in this paper a new filtering technique to reduce the measurement errors on trajectories, speed profiles and acceleration profiles. This technique is based on I-splines; some piecewise polynomials. Then we compare our proposal with a set of filtering technique found in the literature. We use a complete trajectory dataset available within the NGSIM program. As a quality indicator of the various filtering techniques we use velocity distribution, acceleration distribution and jerk analysis for the whole dataset. We also suggest analyzing acceleration standard deviation for each trajectory of the dataset. The main findings are as follows: (1) among the methods compared within this work, the I-splines methods with the action points reduces the most the spikes in the velocity distribution; (2) moreover the I-splines method reduces also the most the percentage of jerk values higher than 15 m/s3 and the percentage of one-second window with more than one sign inversion of the jerk; (3) but in some cases, this method increases the acceleration variability of smoothed trajectories.