The wall being made of reinforced concrete also kills a few use cases. But human detection isn't hopeless in general. You can do pretty well (even gender distinction) by having a physically based model of what you're expecting, e.g. humans, on the ground level instead of at some angle, within a certain range, potentially moving but you can detect heartbeats too, and there are models of how the human body behaves at rest and while moving. (Edit: the assumption of movement really helps a lot since you can diff samples and subtract out all the static bits. So even if clothing attenuates the signal by x%, x < 100, if the human's the only thing in the scene that's moving you're ok.)
(I'm actually impressed/jealous with what they accomplished since it was something I tried and failed to do alone for a school project based on http://ocw.mit.edu/resources/res-ll-003-build-a-small-radar-...)