Air Base from Endlap

The air base is the ground distance the aircraft covers between one exposure and the next, and it comes straight out of the endlap: whatever fraction of the frame must overlap the previous one, the rest is new ground, and that remainder is the base. $B = W_g(1 - p_{end})$. Everything about a flight plan's timing follows from this one number.

Endlap is not sidelap. They are different overlaps, on different sensor edges, doing different jobs, and they are normally set to different values — 75 to 80% endlap along the line, 60 to 70% sidelap between lines. Endlap is what gives you stereo along the direction of travel and what feeds the automatic matcher its long chains of tie points. Sidelap is what stitches adjacent strips into a block. Interchanging the two in a mission planner is the classic error of the trade, and its symptom is a flight that either takes twice as long as quoted or produces a model with holes in it.

Why the numbers are what they are is worth understanding rather than memorising. Classical stereo mapping worked at 60% endlap because a human operator with a stereoscope needed each ground point to appear in exactly two overlapping photographs, and 60% guarantees that with margin for drift. Automated structure-from-motion has a different appetite entirely: the bundle adjustment wants each ground point observed in five to nine images, because redundancy is what makes the solution stable and what lets it detect a bad match. At 60% endlap a point appears in about three frames; at 80% it appears in five. That is the whole reason drone overlaps are so much higher than film-era ones, and it is nothing to do with the sensor.

But more overlap is not free, and the bill comes due somewhere people do not look. Shrinking the base to gain overlap shrinks the base-to-height ratio, and the base-to-height ratio is what conditions the height solution. Push the endlap from 75% to 90% and you have cut the base by 60%, which multiplies your vertical error by two and a half. Overlap buys reliable matching and redundancy; it does not buy height accuracy, and past about 80% it is actively trading one for the other. The cost in frames is also steep, rising as $1/(1-p)$: going from 75% to 90% multiplies the image count along each line by two and a half, with the storage, flight time and processing that implies.

Where more overlap genuinely earns its keep is where matching is hard: forest canopy that moves in the wind, tall buildings that occlude the ground beside them, water, fresh snow, uniform bare soil, and anything photographed in flat overcast light where there is little texture to match on. In those conditions redundancy is the difference between a model and a hole. On well-textured flat ground in good light, 70% is often plenty and the saved battery is real.

Finally, run the equation backwards after a flight. Take the logged exposure positions, difference them to get the base actually achieved, and compare against the frame width: $p_{end} = 1 - B/W_g$. Planned and achieved differ for ordinary reasons — a tailwind stretches the base, terrain rising under the aircraft shrinks the footprint, and a camera that cannot keep up with the commanded interval silently drops frames. That comparison is the fastest way to explain a block that refuses to reconstruct.