You have to make some assumptions about the smoothness of the data, because you lose some information from only having sampled data. See if you can fit the data to a particular model function. If you can't, then you can just estimate the instantaneous velocity = the average velocity at the center of each averaging period. The centers of your averaging periods are between each sample point. If you want the instantaneous velocity at your sample points, then you can take the midpoint between each pair of average velocities. Of course, it's all very approximate.