In maintenance planning, condition-based maintenance (CBM) policies leverage the observation of the current condition of a degrading system to optimize the planning of future maintenance interventions. Researchers often make a strong assumption that remote sensors (potentially coupled with analytic technologies) perfectly capture the health condition of an industrial asset. For many reasons however (e.g., feasibility, cost), such assumption is inexact and the optimization model should take into account the inaccuracy of the monitoring system. This work focuses on a single-item system, continuously but imperfectly monitored, to illustrate how partially observable Markov decision processes (POMDPs) enable this inaccuracy to be considered within the maintenance optimization model. In this framework, the decision-maker minimizes the total discounted cost over an infinite time horizon by taking one maintenance decision at each time step. For the use case we consider, we choose to model three types of interventions: i) preventive maintenance (PM), (ii) corrective maintenance (CM), and iii) perfect inspection (I). A resource constraint is also added to the model to limit the availability of the repair crew. To solve the POMDP, we implement a point-based value iteration (PBVI) algorithm to compute the optimal CBM policy via approximate dynamic programming. In order to guide investment decisions into monitoring technologies, we develop a framework to quantify and compare the value of information provided by different monitoring systems with known average performances. Finally, we analyze how the monitoring quality impacts the structure of the optimal CBM policy.