8.15. Accuracy and Error Sources in Photogrammetry

• Systematic Errors: Due to lens distortion, Earth curvature, tilt. • Random Errors: Due to vibration, atmospheric effects, or human error. • Blunder Errors: Gross mistakes like incorrect GCP location.

Errors in photogrammetry can be classified into three main types: systematic errors, random errors, and blunder errors. Systematic errors are consistent inaccuracies that arise from known sources, such as lens distortion or the curvature of the Earth, which affect measurements uniformly. Random errors are unpredictable and can occur due to factors like vibrations in the equipment or variations in atmospheric conditions at the time of capturing images. Lastly, blunder errors are significant mistakes, such as misplacing a Ground Control Point (GCP) during the planning or execution of a project, leading to major deviations in results.

Think of systematic errors like a ruler that is slightly bent. Every measurement taken with it will be off by the same amount, making your results consistently wrong but predictable. Random errors can be compared to trying to shoot a basketball into a hoop while someone is shaking the hoop; each shot may vary in outcome due to unexpected movements. Blunder errors can be likened to trying to measure the length of a piece of wood but forgetting to account for a few inches – that mistake fundamentally changes the entire project.

• Camera resolution and calibration. • Number and distribution of GCPs. • Image overlap and coverage. • Environmental conditions during image acquisition.

Several factors impact the accuracy of photogrammetric outputs: First, the resolution and calibration of the camera cannot be overstated; a high-resolution camera that is well-calibrated will produce clearer images that can yield more precise measurements. Second, the number and strategic placement of Ground Control Points (GCPs) are crucial; having multiple, well-distributed GCPs across the area of interest improves the reliability of calculated coordinates. Third, adequate image overlap ensures sufficient data is captured for triangulating points accurately. Lastly, environmental conditions such as lighting and weather at the time of image capture can significantly influence the quality of photographs.

Imagine trying to take a photograph of a scenic view. If your camera is blurry (low resolution), the image won't accurately reflect what you're seeing. If you only take pictures in a small area without considering the entire landscape (lack of GCPs), you won't capture a complete picture. Similarly, if you take images only when it’s foggy or too sunny (poor environmental conditions), the quality of your photos will suffer, making it difficult to extract useful information.

• Comparing with ground truth data obtained via total stations or GNSS surveys. • Use of Root Mean Square Error (RMSE) to quantify error. • Visual inspection of orthomosaics and elevation models.

To assess the accuracy of photogrammetric outputs, several techniques can be employed. One common approach is to compare these outputs with ground truth data collected from high-precision instruments like total stations or GNSS surveys. This provides a direct basis for evaluating accuracy. Another method involves quantifying errors using statistical measures such as the Root Mean Square Error (RMSE), which gives an overall estimate of the deviation of measured values from the true values. Finally, visual inspection of orthomosaics and elevation models allows experts to spot any glaring inaccuracies that might not be captured purely through statistical means.

Consider trying to solve a puzzle. You can compare your completed puzzle with the picture on the box to see how closely they match (ground truth comparison). RMSE is like measuring how off pieces are from their expected positions in average — it gives you a sense of how many pieces are misaligned overall. Lastly, visually inspecting the whole puzzle lets you see if any pieces really don’t fit, similar to checking for errors in your photogrammetric model.

--