OBIA for estimating building density

This project employs Object-Based Image Analysis (OBIA) to assess building density patterns, focusing on the Kakuma refugee camp in Kenya. Utilizing high-resolution drone imagery, OBIA segments the buildings which are then analyzed to estimate building density, providing insights into settlement patterns. The findings, presented in a scientific poster format, demonstrate OBIA’s effectiveness in spatial analysis applications within complex environments.

Overview

As part of the course spatial thinking, my colleague and I worked on a project applying Object-Based Image Analysis (OBIA) to quantify building density in the Kakuma Refugee Camp, Kenya. The goal was to understand settlement patterns and demonstrate how spatial thinking techniques can extract meaningful information from high-resolution imagery.

Objectives

The project aimed to:

• Segment individual buildings from high-resolution drone imagery.

• Estimate building density per area and visualize spatial patterns.

Methodology

The workflow began with image preprocessing, where the drone imagery was cleaned, orthorectified, and noise was reduced to improve segmentation quality. Next, segmentation using OBIA was performed by defining scale, shape, and compactness parameters to create meaningful object segments representing individual buildings. During classification, a rule-based approach was applied to identify buildings, with results refined through manual inspection to ensure accuracy.

Finally, in the density analysis step, building footprints were calculated and used to generate density maps, highlighting areas of high and low building concentration across the study area.

Results

The scientific poster below summarizes the work carried out. Using OBIA, we successfully segmented and classified buildings across the study area, producing building density maps that clearly highlight spatial patterns. The analysis revealed zones of both high and low building concentration, providing insights into settlement structure and potential areas for planning or humanitarian support.

Conclusion

Given the very high-resolution imagery, there is a clear need to explore deep learning methods for more efficient building extraction. In this project, building density was calculated using a rigid 1 km × 1 km grid, which is relatively coarse compared to the actual building distribution. Using more flexible shapes, such as hexagons, could better capture spatial variability. Future work should consider adaptive density metrics and advanced segmentation techniques to produce more realistic representations of settlement patterns.