Monitoring air quality in real-time with the help of expensive monitoring station is expensive. IoT-based sensors might be a cost-effective solution. However, sensors are resource-constrained. This paper proposes a framework consisting of physical and virtual sensors for real-time pollution monitoring. In this framework, virtual sensors are created to overcome the limitations of physical sensors, such as calibration, loss of information, and lack of accuracy. Virtual sensors can predict missing data that may be lost due to an error in the devices or during transmission. We propose to divide the entire area into small zones and strategically place physical sensors in these zones so that no two adjacent zones have a similar set of sensors. Virtual sensors are used here to predict the data from sensors that are not present in a zone using the data from adjacent zones. In this way, number of physical sensors in a specific geographic area is reduced, and a cost-effective framework is developed. Advanced machine learning techniques may be implemented on top of the framework to predict the quality of air and the concentration of pollutants.