The issue might not lie with your drone; it could be what took place to your images after the flight.
One of the most annoying situations in this field occurs when the flight proceeds without any problems, the pilot sticks to the planned procedure, and there are no obvious issues on the ground — yet the processing software produces a model riddled with gaps, a blurry orthomosaic, or a DSM containing spikes that don't belong. The natural reaction is to place the blame on the drone or the software; typically, however, the actual cause occurs at some point between takeoff and the moment you press "process", in an aspect which no one was keeping a close enough watch on.
The Usual Suspects
Very little image overlap. This is the main reason why models fail. The standard minimum level is 75 to 80 per cent front overlap and 65 to 75 per cent side overlap — yet this minimum should be increased, frequently reaching 90 per cent or more, especially when the terrain is featureless, involves complex structures, or has changing elevation. Failure to achieve sufficient overlap not only lowers the quality but, below a certain threshold, prevents the software from finding enough matching features between the images to reconstruct the scene. We've gone deep on exactly this question before — how many overlap photos you actually need, broken down by terrain type and mission type.
Motion blur. When the degree of blur goes beyond that of a single ground pixel, the quality of the reconstruction begins to degrade — and the annoying thing is that the individual images usually appear satisfactory during a quick field check. The blur only becomes apparent when you look at the final processed image. The typical causes are strong wind, flying faster than the shutter speed, and consumer-grade drones using rolling-shutter cameras.
Rolling shutter distortion. It is related to blur but not the same; since a rolling shutter camera takes the image line by line rather than capturing it all at once, any movement that occurs during the capture process causes geometric distortion, not just blur. Cameras used in consumer drones are more susceptible to this effect than global-shutter survey-grade sensors.
Poor lighting. During very early morning, in the late evening, and under harsh midday sunlight, each situation has its own set of problems — long shadows make it difficult to match features and, in low light, longer exposures are necessary, thus increasing the risk of blur. It is usually the overcast days, not the worst, that are most suitable for carrying out photogrammetry.
Poor or wrong distribution of GCPs. While the number of points is important, the way they are distributed is equally so — research in this area shows that if the reliability of the GCPs is high, then it is better to focus on the number; if the reliability is moderate, then it is more important to spread the points across the site than to add more points; and when the reliability is low, a distribution with points concentrated at the edges and in the centre performs better than a uniform distribution. A tight cluster of GCPs in one corner of the site does not benefit the rest of the model, no matter how many points there are.
Vegetation and repetitive surfaces. Water, uniform crop fields, and dense, homogeneous vegetation are particularly problematic for feature-matching algorithms since there is usually not sufficient visual variation between adjacent images for the software to establish a match. This is directly reflected in the processing reports in the form of images that are labelled "uncalibrated".
Poor camera calibration. If a camera model is not calibrated or is only poorly calibrated, it will introduce systematic errors over the whole dataset, not just in a few images — and this type of error is one that is easy to overlook until well into the processing stage.
Terrain variation. A flight plan which has been adjusted for flat ground will not automatically function when the area in question involves a significant change in elevation — the overlap that was sufficient on level ground can effectively fall below the required threshold on slopes and along ridgelines, since both the ground sample distance and the viewing angle change with elevation.
Incorrect coordinate systems. Combining a height-above-ground figure with software that is based on orthometric height is a particular and well-documented kind of failure; it can cause the software to calculate zero overlap and prevent the camera model from being calibrated at all, even though the images themselves are unproblematic.
Processing settings which do not match the capture settings. This can happen if the RTK/GCP settings are incorrectly configured, if the processing quality setting differs from what the capture operation actually supports, or simply because default settings are used on a dataset that required more deliberate settings — in all these cases a good flight can end up with a poor result.
There are not enough tie points. This is a direct result of low overlap or poor feature matching — if there are too few tie points between the images then the geometric solution will be weaker and less accurate, even though the final output may appear fairly good.
Orthomosaic seams and distortions. Seams that are visible where adjacent image blocks fail to blend properly are generally a sign rather than an independent issue; they are usually due to problems with overlap, the distribution of GCPs, or calibration that occurred at an earlier stage.
DEM/DSM artifacts. Spikes, pits, and spurious elevation features in a surface model are usually the result of objects having moved during the capture process, of water surfaces, or are due to the same kinds of vegetation and reflectivity problems that cause feature-matching difficulties elsewhere in the pipeline.
Where Each of These Actually Lives in the Workflow
It helps to place these failure points against the actual pipeline, because "processing failed" almost always traces back to an earlier stage:
Flight Planning → the overlap percentages, the altitude, and the strategy for placing the Ground Control Points (GCPs) are determined at this stage, prior to any image being taken. This is directly linked to the pre-survey checklist, since the flight time calculations and the visibility of the GCP targets are decisions that must be made during the planning stage and not corrected during the processing stage.
Image Acquisition → when the shutter is triggered, motion blur, rolling shutter effects, and the lighting conditions are fixed, and nothing that comes after can fully recover any details that were not originally captured.
Quality Control → this is the stage most frequently omitted when time is tight. A rapid examination of a small number of images is not equivalent to checking for blur, exposure problems, and coverage gaps before departing from the site.
Georeferencing → this is where the identification of GCPs, consistency of the coordinate system, and the assumptions regarding height reference are all settled, and it is at this stage that the problems relating to the discrepancy between orthometric and ellipsoidal heights and the distribution of the GCPs appear.
Photogrammetric Processing → feature matching, the generation of tie points, and optimisation of camera calibration happen here, which is the reason why problems or warnings related to vegetation, repetitive surfaces, and earlier calibration issues appear at this stage.
Orthomosaic/DSM/DTM Generation → the production of seams, blending artifacts, and spikes in the surface model happens here, but these are almost always later consequences of a choice made at an earlier stage.
Accuracy Assessment — it's in this stage that you can actually detect a problem with the GCP distribution or georeferencing using concrete figures, not just by visual inspection, just as you would when proving to a client that a certain GSD accuracy claim has been met.
Final GIS/CAD Deliverables — once a problem arises at this stage, it has become the most costly point in the process to trace back to and correct, since the cause may have originated in any of the six previous stages.
The Actual Takeaway
These failure points are by no means unusual; each one has been well recorded, and the majority of them have a known solution. The point that should be remembered is that when processing fails it is very seldom due to a problem in processing — it is almost always the result of a flaw in planning, in acquisition, or in quality control which only comes to light at the processing stage. It is better to detect such a problem at that stage than not to detect it at all, but it is the detection of the problem two or three stages earlier that actually stops the need for a redo.
If your team is frequently encountering these problems and doesn't know where in the pipeline the issue is actually arising, then this is precisely the kind of diagnostic work that drone imaging and processing services are designed to carry out — identifying the real cause of a defective deliverable rather than simply reprocessing the same dataset and hoping for a different outcome.
