Motion capture through inertial sensors is becoming popular, but its accuracy to describe kinematics during changes in walking speed is unknown. The aim of this study was to determine the accuracy of trunk speed extracted using an inertial motion system compared to a gold standard optical motion system, during steady walking and stationary periods. Eleven participants walked on pre-established paths marked on the floor. Between each lap, a 1-second stationary transition period at the initial position was included prior to the next lap. Resultant trunk speed during the walking and transition periods were extracted from an inertial (240 Hz sampling rate) and an optical system (120 Hz sampling rate) to calculate the agreement (Pearson’s correlation coefficient) and relative root mean square errors between both systems. The agreement for the resultant trunk speed between the inertial system and the optical system was strong (0.67 < r ≤ 0.9) for both walking and transition periods. Moreover, relative root mean square error during the transition periods was greater in comparison to the walking periods (>40% across all paths). It was concluded that trunk speed extracted from inertial systems have fair accuracy during walking, but the accuracy was reduced in the transition periods.