You can use smoothing functions such as 'smoothdata' to create a moving average filter with a specified window length. The following documentation link explains this in detail:
However, it is important to note that when implementing this in a practical setting, you will need to buffer the data which is being streamed in and then apply this smoothing to that data. This will introduce a delay in the measurements of the velocity as the moving average filter will need a set of values from time instant 1 to say time t.
Another useful approach to handle such scenarios is to build an estimator using a Kalman filter. The estimator will estimate the velocity from the distance measured and this will perform well even with measurement noise or at least converge to good measurements over time. Since the relationship between velocity and distance is known you can design a set of equations which will model this behavior.
The following documentation links are a good starting point to understand Kalman filters and tools available for you to design an estimator:
