There are two issues you need to address. First, what to do with the traffic you send to MPLS. Second, what happens to the traffic as it exits MPLS.
For the former, much as the other posters have also noted, a hierarchical shaper, with the parent's shaper configured for 18 Mbps, and the child's video class provided 16 Mbps, would often be the norm. If the video is not real-time, an ordinary CBWFQ class reservation of the necessary bandwidth usually works fine. If the video is real-time, e.g. video conferencing, LLQ is often the choice, but at these bandwidth levels, it's often not necessary.
e.g. (pseudo config)
policy-map parent
class class-default
shape 18 Mbps
policy-map child
class video
bandwidth 16 Mbps (or priority 16 Mbps)
class class-default
no fair-queue (FQ, on most platforms, seems to disturb non-LLQ bandwidth reservations)
fastethernet 0/0
service-policy output parent
The latter issue, what happens at MPLS egress, can be just as important. If your MPLS environment is multi-point, it's crucial too. If there are only two sites, your outbound policy, alone, might be all you need, but you still need to confirm what the MPLS provider does. For instance, they might rate-limit the MPLS egress bandwidth, and if their burst intervals differ from your shaper's, even though you both are configured for 18 Mbps, they might still drop some of your traffic.
If you're working with multi-point MPLS, then you need to consider how to obtain the bandwidth reservation you need within that environment which is very dependent on what the MPLS vendor offers. The latter is usually accomplished by usage of some type of CoS/QoS marking, for which the MPLS provider guarantees some service level for particular markings.
