View-invariant recognition of body pose from space-time templates