Integrated Analysis of Ground Level and Aerial Image Data

Integrated analysis of ground level images with aerial image data is a new area of computer vision research which is gaining traction due to the availability of high volumes of crowd-sourced and open data that are being collected worldwide. Novel problems such as geo-location from a single or a sequence of ground level images and use of geospatial context in object recognition are active research topics. Geo-tag metadata from image data as found in social media can be utilized to extract aerial imagery surrounding the location where an image was acquired. While images found in social media repositories provide close up details of structures from ground view, aerial images have other desirable properties such as known resolution, geometry etc. The approach adopted in this project leverages features from both ground level and aerial imagery that are similarly geo-located to improve the classification of visual scenery around that place. Image feature extraction and machine learning methods are being evaluated to test the efficacy of this approach in Volunteered Geographic Information. This work can improve scenery classification for use in large scale image indexing and enhance the use of crowd-sourcing as affordable and practical means of filling information gaps in land use information.