Algorithm 2 Offline algorithm for sensor scheduling.

Input:${\pi }_l$, $A_{\mathcal{G}}$, $E^m$, $T_{l,h}^j$   From CRM KB:  ${\pi }_l$: the choices of groups of sensors $\mathcal{G}$ for each of the requested L contexts  $A_{\mathcal{G}}$: the accuracy of group $\mathcal{G}$  $E^m$: the energy consumption by each sensor $S^m$   From User Behavior:  $T_{l,h}^j$: the time limits of the context states $x_l^j$ Output:$a_{\mathcal{G}}^i$, $E_{\mathcal{G}}^j$, $D_{\mathcal{G}}^j$  $a_{\mathcal{G}}^i$: the sensing schedules for the sensor groups in ${\pi }_l$ recognizing each context.  $E_{\mathcal{G}}^j$: the energy consumption for each group of sensors $\mathcal{G}$ recognizing context state $x_l^j$  $D_{\mathcal{G}}^j$: the delay of each sensor group $\mathcal{G}$ in recognizing context state $x_l^j$ 1:for each sensor group $\mathcal{G}\in {\pi }_l$ do 2:  Calculate $E_{\mathcal{G}}$ as the sum of $E^m$ of $S^m\in \mathcal{G}$ 3:  for each context state $x_l^j$ do 4:   for each ($\alpha ,\beta$) pair do 5:    Run Viterbi algorithm to determine sensing schedule for $\mathcal{G}$ in recognizing state $x_l^j$ 6:    Compute the objective function value according to (12) 7:    Track the resulting energy consumption, delay, and the ($\alpha ,\beta$) pair for the derived sensing schedule. 8:   end for 9:   Store the sensing schedule with the minimum objective function value according to (12) and the associated energy consumption ($E_{\mathcal{G}}^j$), delay ($D_{\mathcal{G}}^j$), and ($\alpha ,\beta$) pair in the Look-Up Table (LUT) 10:  end for 11:end for