The use of pixel-based algorithm for automatic change detection of 3D Building from Aerial and Satellite Imagery : Erbil city as a case study

Abstract EN

Detection of Three Dimension (3D) changes and monitoring urban areas using photogrammetric and remotely sensed data is becoming an important research topic for monitoring the city development, disaster assessment, earthquake monitoring, and updating geo-database.

In practice, identifying 3D changes manually in urban areas, specifically when dealing with a large number of buildings is considered to be a very time-consuming task, in such cases, automatic 3D change detection is considered to be very cost-effective.

This paper presents an algorithm which is based on using pixels differencing to automatically detect 3D change of the buildings that occurred in the selected study area (Erbil city) for the periods from 2012 to 2017 by subtracting two digital surface models (DSMs) generated from two different datasets that has been captured from two different sensors.

The first dataset is from stereo aerial imagery captured in (2012) and the second dataset is based on Very High Resolution (VHR) stereo satellite imagery captured in (2017).

The proposed method is applied to three study areas (Ankawa, Dream city and 32 park) in Erbil city.

Prior to applying change detection algorithm, the vertical accuracy of the DSMs is checked, through field point measurements by Differential GPS (DGPS).

The presented work in this article deals with building change detection.

The changes that refer to differences in size and shape of buildings are considered significant, while changes in other urban objects, such as roads, ground and vegetation, are considered insignificant and needs to be removed.

Through some post-processing steps that performed to preserve only the real changes and eliminate the virtual ones.

The outcome of this study revealed that for study area one (Ankawa), 105 out of 157 changed buildings are detected correctly.

While in the study area two (Dream city) 74 out of 106 changed buildings are detected correctly, and for study area three (32 park) the result was more accurate and 28 out of 31 changed buildings are detected correctly.