Autonomous rovers meant for exploring unknown terrains use multiple sensors for mapping the environment surrounding it. The quality of the map generated affects the ease and feasibility of exploration. In this paper we present an exploration technique which uses monocular vision to create a metric map of the environment for navigation. The navigation behaviour relies on frontier based exploration. We propose modifications to improve the computational complexity of the algorithm required for exploration. These modifications are mainly in the areas of identifying the sub goals marking the unexplored area boundary and choosing the subsequent goal for navigation in an efficient manner. Also, the map generation scheme has been improved by a combination of methods which improve the ground plane extraction and the map update model. We present experimental results demonstrating the exploratory behaviour on our experimental rover in environments of varying obstacle distribution.