http://www.textronsystems.com/intel-hub/durham-case-study
Textron Systems used a city boundary shapefile for guidance to define a test area of five square kilometers within city limits, and then clipped out a stack of sample data from Nearmap’s online holdings that ranged from 2014 to 2016. Next, Textron Systems replicated the GIS Group’s existing workflow model on the 2014 dataset, then began preliminary tests to analyze results and modify the workflow along the way. To assess quality and accuracy, random sampling was chosen as the quality control method to validate the results. To test the fidelity of the feature extraction model, the model was applied to the data from 2016 to ensure that it was consistent with the 2014 results.
The final test results yielded 190 new impervious surface polygons identified through automated feature extraction. To verify accuracy, 50 of the 190 results were randomly selected for visual inspection and then individually marked in the attribute table as either correct, incorrect, possible or other. The results of the quality assessment yielded the following statistics.
- 37 correct results for new impervious areas
- 7 possible results for new impervious areas that would need further verification
- 5 incorrect results
- 1 water feature
Based on these results, Public Works could expect to achieve an approximate accuracy rate of 85-90 percent using Nearmap imagery and Feature Analyst.
To measure performance, the total quality control time, or the time it took to perform random sampling which was 15 minutes, was added to the total extraction time of one minute and then rounded up to 20 minutes per five square kilometers. This averaged out to be approximately four minutes per square kilometer for a preliminary result. Based on this proof of concept, the city was pleased with the results and decided to move forward with its transition to Nearmap data.
These results demonstrate the potential of automated feature extraction to accomplish a task that would normally take days or weeks in mere minutes. In addition, it provides a real-world example of a collaborative relationship between the imagery vendor, software developer and end user to develop a robust solution for all parties involved. Each player serves as an expert in their own domain, while at the same time learning from the others to create a capability that can be built upon over time.
