Interaction with objects is crucial for robots operating in human environments. In this work, we are looking at the problem of utilizing semantic knowledge collected by the robot in order to permit such tasks as active visual object search in realistic large-scale envrironments. In this case, the active visual object search is defined as a problem of localizing an object in an unknown 3D environment by executing a series of actions with the lowest total cost. The cost is defined in terms of time it takes to complete the task or distance traveled. We make use of a hierarchical planner combined with our semantic mapping algorithm. The resulting system uses the collected knowledge to explore further and direct the search towards places that are more likely to contain the target objects. The system is capable of trading exploration vs exploitation during the search process.