In this project, we consider the problem of cross-view image-based ground-to-aerial geo-localization: estimating the GPS location of a query ground view image by finding corresponding matching images in an aerial/satellite view image database, or vice versa. The major challenge is the radical change of viewpoint and different light conditions as well as seasons. In our project, we examine recent studies in deep-learning-based approaches for cross-view matching. Specifically, we train and test these methods on a dataset collected on photorealistic simulator AirSim. Other than working with panoramic ground view images and satellite imagery, these methodologies are modified to work well with non-panoramic ground view images and aerial images taken from smaller altitude. We also employ particle filtering technique to accomplish the geo-localization task using the results from cross-view image matching. We demonstrate the effectiveness of our methods with intensive experiments.