Change Detection Based on Object-based Features from Multi-temporal, Multi-source Orthoimages and Digital Surface Models

Unmanned Aerial Vehicle (UAV) attached with a non-metric camera is becoming a popular platform for acquisition of aerial images. Drone mapping has become a cost-effective approach for acquiring temporal images over large geographic areas. With a well-planned flight mission, which considers the image overlap, camera angle, and flight altitude, one can rapidly produce geo-referenced orthoimages and digital surface models (DSMs). These two types of data are fundamental for Earth observation. Also, using multi-temporal geospatial data can be beneficial in understanding landscape change.

In this study, multi-temporal orthoimages and DSMs produced by fix-wing aerial photogrammetry and UAV multi-view stereoscopic technique were employed for change detection. Object-based features extracted from orthoimages and DSMs were analyzed. A statistical tool, called estimation of scale parameter (ESP), was used in estimating the appropriate scales for the multi-resolution segmentation. Also, a feature based on rectangular fitting was employed to discriminate natural changes from artificial changes. The spectral features based on the differences in orthoimages and surface heights were both employed. The source of orthoimages were collected by an UltraCam camera attached in a fix-wing aircraft and sensfly S.O.D.A camera attached in an eBee drone. The results show where the changes are, and identifies what the changes are. For example, the geographic coverage and volume caused by a landslide event are measured; the newly-built buildings are identified; and where the lost trees are found.