Confronting passive and active sensors with non-Gaussian statistics

Area of Science:

• Geomatic Engineering
• Photogrammetry
• Remote Sensing

Background:

• The geomatics field faces intense competition, driving innovation in remote sensing technologies.
• Terrestrial photogrammetry, enhanced by computer vision, is emerging as a competitor to laser scanning.
• Evaluating the performance of different remote sensing systems for architectural object modeling is crucial.

Purpose of the Study:

• To compare Digital Surface Models (DSM) generated by passive (digital camera) and active (terrestrial laser scanner) systems for architectural objects.
• To assess the suitability of Gaussian statistics and Least Squares for data with asymmetrical errors.
• To investigate the efficacy of non-parametric statistical methods for such datasets.

Main Methods:

• Development and application of new photogrammetry software.
• Design and execution of an experimental test comparing digital camera and terrestrial laser scanner data.
• Statistical analysis using Gaussian methods (Least Squares) and non-parametric methods (Median Absolute Deviation, Biweight Midvariance).

Main Results:

• Good agreement was observed between DSMs from both sensors, even with significant data asymmetry.
• Standard deviation, based on Gaussian statistics, proved inadequate for assessing data with asymmetry.
• Median Absolute Deviation and Biweight Midvariance provided more appropriate accuracy estimations for asymmetrical datasets.

Conclusions:

• While both passive and active remote sensing systems can yield comparable DSMs for architectural objects, data characteristics influence statistical analysis.
• Gaussian statistics and Least Squares are not ideal for datasets with asymmetrical gross errors.
• Non-parametric statistical measures are recommended for robust accuracy assessment in such scenarios.