Photogrammetry and drone surveys

Civil Engineering (B.E./B.Tech) · Surveying · free study notes for exams, placements and interviews

In short

Photogrammetry is the science of obtaining reliable measurements of position, shape and size of objects from photographs, chiefly from overlapping aerial images that are viewed or processed in stereo. It converts a camera into a measuring instrument, so large areas are mapped far faster than by chain, total station or GPS traversing. Drones (UAVs) have made the technique routine in India: SVAMITVA rural property mapping, NHAI and Railway alignment surveys, mining volumetrics, irrigation and smart-city topographic surveys. It carries marks in Surveying/Geomatics papers and in GATE (CE) under "photogrammetry and remote sensing", and drone-survey skills are now an explicit hiring requirement for junior engineers in survey, infrastructure consultancy and GIS firms.

Explained

Fundamentals, types and basic geometry

Definition: measurement from photographs. Metric photogrammetry extracts coordinates, distances, areas, volumes and elevations; interpretative photogrammetry identifies features (land use, crops, damage).
By camera station: terrestrial (camera on ground, close-range work on bridges, façades, heritage), aerial (aircraft/drone), space (satellite stereo imagery).
By method: analogue (optical-mechanical plotters), analytical (computed from measured photo coordinates), digital/softcopy (image matching on computers) — current practice is wholly digital.
Camera terms: focal length f, fiducial marks defining the principal point (p), lens distortion. A camera whose f and principal point are calibrated is a metric camera; drone cameras are non-metric and are self-calibrated during processing.
Three special points on a photo: principal point (foot of perpendicular from lens), nadir point (vertically below the lens), isocentre (bisects the tilt angle). They coincide on a truly vertical photograph.
Photograph classes: vertical (optical axis within ~3° of vertical — used for mapping), tilted (3°–30°), oblique — low oblique (no horizon) and high oblique (horizon visible).
Orientation: interior orientation recovers the camera's internal geometry (f, principal point, distortion); exterior orientation gives six parameters per photo — X₀, Y₀, Z₀ and the rotations omega, phi, kappa (tilt, tip, swing). Relative orientation fixes one photo relative to its pair; absolute orientation scales and levels the model to ground coordinates using control.
Collinearity condition: the object point, the perspective centre and its image lie on one straight line; two collinearity equations per image point are the backbone of bundle block adjustment, which solves all exterior orientations (and camera calibration) simultaneously by least squares.
Photo versus map: a photograph is a central (perspective) projection, so scale varies with ground height and displacements occur; a map is an orthogonal projection at a constant scale. Removing this difference produces an orthophoto.

Scale, relief displacement and tilt

Scale of a vertical photograph S = f/(H − h): H is flying height above datum, h the ground elevation. Scale therefore changes point to point over undulating ground.
Datum scale = f/H; average scale uses mean ground elevation. Scale can also be found from a known ground distance: S = photo distance / ground distance.
Relief displacement: points above the datum are pushed radially outward from the principal point, points below pulled inward; d = r·h/H. It is zero at the principal point and maximum at the corners — which is why tall buildings "lean outward" in a photo.
Relief displacement is a nuisance for mapping (removed by orthorectification using a DEM) but useful: it gives object heights from a single photograph.
Tilt displacement is radial from the isocentre and is positive (outward) on the lower side of the photo; it is removed by rectification. Modern digital workflows absorb tilt in the exterior orientation.
Flight errors: drift (aircraft pushed sideways by wind, flight line shifts), crab (aircraft yawed so photo edges are not parallel to the flight line) — both reduce effective overlap and are countered by extra sidelap.

Stereoscopy and parallax

Stereoscopic vision needs two photographs of the same ground from different stations; the overlap area viewed through a stereoscope (mirror or pocket) gives a three-dimensional stereo model.
Air base (B): the ground distance between two successive exposure stations; photo base (b): its image on the photograph.
Parallax is the apparent shift of a point between the two photographs, measured parallel to the flight line: p = x − x′. Absolute parallax of a datum point p = B·f/H. Higher points have larger parallax.
Parallax difference (Δp) between two points gives their height difference: h = H·Δp/(p + Δp), measured with a parallax bar (stereometer) and a floating mark.
Floating mark principle: two half-marks fuse into a single mark that appears to float in the model; when it rests exactly on the terrain, the parallax reading gives the elevation. This is how contours were traced in stereo-plotters and is still how operators check a digital surface.
Vertical exaggeration: stereo models appear steeper than reality because the base-to-height ratio of the photography exceeds that of human eyes; useful for interpretation, misleading for slope judgement.
Y-parallax (shift perpendicular to the flight line) is an error from imperfect relative orientation and must be cleared before measurement.

Flight planning for aerial photography

Fix mapping scale, contour interval and accuracy required; from these choose the photo scale and hence flying height H = f/S.
Flying height from contour interval: H = C × contour interval, where C-factor is an instrument/process constant (commonly taken as about 1,000–2,000; it varies with equipment, so use the value given in the question or specification).
Overlaps: longitudinal (forward) overlap ≈ 60% — needed so every point appears on at least two photos; lateral (side) overlap ≈ 30% between adjacent strips. Drone/SfM work uses much higher values (75–85% front, 60–75% side).
Compute ground coverage per photo from the format size, then air base, flight-line spacing, number of photos per strip, number of strips and total photographs (add 2 extra photos at each strip end).
Exposure interval T = B/V (V = ground speed); the intervalometer or flight-control software triggers accordingly.
Fly strips parallel to the long dimension of the area to reduce turns; fly in the season and hour that give clear skies, low haze and a sun elevation above about 30° so shadows are short.
Provide ground control before or after flying, and flight-plan maps, index mosaics and a flight log as deliverables.

Drone (UAV) surveying and the SfM workflow

Platforms: multirotor (vertical take-off, hovers, best for small sites, façades, stockpiles; endurance 20–40 min), fixed-wing (covers large areas and corridors efficiently, needs launch/landing space), VTOL hybrid (combines both), and UAV-LiDAR for vegetated terrain.
Payloads: RGB mapping camera (most common), multispectral (crop/irrigation), thermal (solar plants, leak detection), LiDAR (penetrates canopy to give bare-earth DTM).
Positioning: standard GNSS on board gives metre-level image positions; RTK/PPK drones tag each image to centimetre level against a base station or CORS, cutting the number of ground control points drastically — but ground checkpoints are still mandatory to prove the accuracy.
Structure from Motion (SfM) / digital photogrammetry pipeline:
1. Image acquisition per flight plan (fixed GSD, high overlap, consistent exposure; add cross-strips and oblique images for tall structures).
2. Feature detection and matching across overlapping images → tie points.
3. Bundle block adjustment with GCPs → exterior orientation, self-calibration, sparse point cloud.
4. Dense point cloud generation by multi-view stereo matching.
5. Classification and filtering → DSM (surface, includes trees and buildings) and DTM (bare earth).
6. Orthomosaic by orthorectifying and mosaicking; contours, breaklines, volumes, CAD/GIS deliverables.
Ground control points (GCPs): well-defined, painted or targeted marks (typically 0.6–1.0 m "L" or chequer targets) coordinated by DGPS/total station, spread around the perimeter with some in the interior and at height extremes. A common practical rule is a minimum of 5 GCPs plus extra for every few hectares, with independent checkpoints not used in the adjustment.
Typical accuracy: horizontal RMSE about 1–2 × GSD, vertical about 2–3 × GSD, provided the control is good. Poor GCP distribution produces "doming" — a systematic bowl in the surface caused by weak self-calibration.
Limitations: cannot see under tree canopy or water; weather and wind dependent; poor results on featureless surfaces (sand, fresh snow, still water, uniform roofs); legal restrictions near airports and sensitive sites; battery endurance.

Indian regulations, standards and practice

Drone Rules, 2021 (Ministry of Civil Aviation), which replaced the UAS Rules 2021 and have been amended since, govern civil drone operation. They abolished many earlier approvals and reduced the number of forms and fees.
Weight categories: nano (≤250 g), micro (>250 g–2 kg), small (>2–25 kg), medium (>25–150 kg), large (>150 kg). Most survey drones are micro or small.
Registration: every drone, including nano, must be registered on the Digital Sky portal and carry a Unique Identification Number (UIN) — nano drones are exempt from type certification and from the pilot certificate, not from registration; manufacturers need type certification under the Certification Scheme for Unmanned Aircraft Systems (CSUAS).
Airspace map: the Digital Sky interactive map divides Indian airspace into green, yellow and red zones. No prior flight permission is required in a green zone up to 120 m (400 ft) AGL; yellow zones (controlled airspace, and the area around airports) need ATC/ MoCA permission; red zones are prohibited without Central Government clearance.
Pilot: a Remote Pilot Certificate (RPC) issued through a DGCA-authorised Remote Pilot Training Organisation (RPTO) is required (nano, and micro flown non-commercially, are exempt).
Other standing requirements: daylight, visual line of sight (VLOS) operation; BVLOS only with specific permission/under the regulatory sandbox; no dropping of payloads without permission; insurance as notified. Real-time tracking/"No Permission–No Take-off" provisions have been phased and amended — confirm the current notification before an operation, as this area changes frequently.
Geospatial side: the Guidelines for acquiring and producing Geospatial Data, 2021 (DST) liberalised map making by Indian entities, and the National Geospatial Policy, 2022 sets targets for high-resolution national topographic and elevation data. Survey of India remains the national mapping agency and supplies CORS network corrections.
Programmes and mandates: SVAMITVA (Ministry of Panchayati Raj with Survey of India, from 2020) drone-maps inhabited rural *abadi* land at roughly 5 cm GSD for 1:500 property cards across lakhs of villages; NHAI/MoRTH use drone-LiDAR and photogrammetry for DPR corridor surveys and monthly construction monitoring; Indian Railways uses drones for alignment and asset surveys; the Indian Bureau of Mines and State mining departments require periodic drone surveys of leases for volumetric reconciliation. There is no single IS code for drone survey accuracy — specifications normally quote ASPRS-style RMSE limits or a departmental technical specification.

Key points

Photogrammetry - obtaining metric information (position, shape, size) about objects from photographs.
Central projection - a photograph has one perspective centre, so its scale varies with ground elevation; a map does not.
Scale - S = f/(H − h); H is above the datum, h is the ground height at that point.
Relief displacement - d = r·h/H, radial and outward from the principal point for points above datum; gives heights from a single photo.
Tilt displacement - radial from the isocentre; principal point, nadir point and isocentre coincide only on a vertical photo.
Overlap - about 60% forward and 30% lateral in classical aerial work; 75–85% and 60–75% for drone/SfM.
Air base B - ground distance between successive exposures = (1 − forward overlap) × ground length of one photo.
Parallax - p = B·f/H; parallax difference Δp gives height, h = H·Δp/(p + Δp).
Floating mark - fused half-marks set on the terrain in a stereo model to read elevation.
Stereoscopic coverage - every ground point must appear on at least two photographs, hence 60% forward overlap.
Drift and crab - wind-induced lateral shift and yaw of the aircraft; both eat into effective overlap.
Interior / relative / absolute orientation - camera geometry, pair geometry, then scaling and levelling to ground coordinates.
Bundle block adjustment - simultaneous least-squares solution of all exterior orientations using collinearity equations and GCPs.
GSD - ground sample distance = (pixel size × H)/f; the single most important drone planning parameter.
DSM vs DTM - surface including objects versus filtered bare earth; contours and earthwork come from the DTM.
Orthophoto - photograph rectified for tilt and relief so it can be measured like a map.
RTK/PPK - centimetre image geotags that reduce GCP count but never remove the need for independent checkpoints.
Drone Rules 2021 - UIN on Digital Sky, Remote Pilot Certificate, green zone flying up to 120 m AGL without prior permission.

Formulas and examples

Scale of a vertical photograph: S = f / (H − h) = photo distance / ground distance
f = focal length (m), H = flying height above datum (m), h = ground elevation above datum (m)
Ground length covered by one photo: Lg = l × (H − h) / f ; l = photo format side (m)
Relief displacement: d = r·h / H ; r = radial distance of the image point from the principal point, d and r in photo units, h and H in the same ground unit
Air base: B = (1 − Pl) × Lg ; flight-line spacing: Sy = (1 − Pw) × Wg ; Pl = forward overlap, Pw = side overlap (fractions)
Number of photographs: N = A / [ (1 − Pl)(1 − Pw) × Lg × Wg ] ; A = area to be covered
Exposure interval: T = B / V ; V = ground speed (m/s), T in seconds
Absolute parallax: p = B·f / H ; height from parallax difference: h = H·Δp / (p + Δp)
Flying height from contour interval: H = C × contour interval ; C = C-factor of the plotting process
Ground sample distance: GSD = (pixel size × H) / f ; required height: H = GSD × f / pixel size
RMSE = √( Σ(residual²) / n ) ; residuals at independent checkpoints, in metres

Example 1 — Aerial flight plan. Camera f = 152 mm, format 230 mm × 230 mm; area 100 km² (10 km × 10 km) at mean ground elevation 400 m above MSL; aircraft flies at 2,500 m above MSL at 200 km/h; Pl = 60%, Pw = 30%.

Scale = 0.152/(2500 − 400) = 0.152/2100 = 1/13,816, i.e. 1:13,816.
Ground side of one photo = 0.23 × 13,816 = 3,177.6 m; ground area per photo = 3.1776² = 10.10 km².
Air base B = 0.40 × 3,177.6 = 1,271 m; flight-line spacing = 0.70 × 3,177.6 = 2,224 m.
Effective area per photo = 0.40 × 0.70 × 10.10 = 2.83 km² → N = 100/2.83 = 35.4, say 36 photographs before end-of-strip extras.
Strips = 10,000/2,224 = 4.5 → 5 strips; photos per strip = 10,000/1,271 = 7.9 → 8, plus 2 extra at each end = 12; total = 5 × 12 = 60 exposures.
Ground speed = 200 × 1000/3600 = 55.56 m/s → exposure interval T = 1271/55.56 = 22.9 s.

Example 2 — Height of a chimney from relief displacement. On a vertical photograph taken from 1,200 m above the chimney base, the image of the chimney top is 85.0 mm from the principal point and the top-to-base image displacement is 2.5 mm.

h = d·H/r = 2.5 × 1200 / 85.0 = 35.3 m. (Both d and r are photo distances, so units cancel.)

Example 3 — Height difference from parallax. Flying height 2,000 m above datum, air base 1,200 m, f = 152 mm. Parallax bar readings for two points differ by Δp = 2.40 mm.

Datum parallax p = B·f/H = 1200 × 0.152/2000 = 0.0912 m = 91.2 mm.
Height of the higher point above datum, h = H·Δp/(p + Δp) = 2000 × 2.40/(91.2 + 2.40) = 4800/93.6 = 51.3 m.

Example 4 — Drone mission planning. Camera: 20 MP, image 5,472 × 3,648 px, sensor 13.2 mm × 8.8 mm, f = 8.8 mm. Site 50 ha (0.5 km²); target GSD 2.0 cm; front overlap 80%, side overlap 70%; cruise speed 5 m/s.

Pixel size = 13.2/5472 = 2.412 × 10⁻³ mm = 2.412 µm.
Required height H = GSD × f / pixel = 0.020 × 0.0088 / 2.412 × 10⁻⁶ = 73.0 m AGL (well inside the 120 m green-zone limit).
Footprint = 5,472 × 0.02 = 109.4 m (across track, flying with the long side across) × 3,648 × 0.02 = 73.0 m (along track).
Flight-line spacing = 0.30 × 109.4 = 32.8 m; trigger distance = 0.20 × 73.0 = 14.6 m → interval = 14.6/5 = 2.9 s.
Effective ground area per image = 14.6 × 32.8 = 479 m² → images = 500,000/479 = 1,044, say ~1,100 with cross-strips.
Line length 1,000 m per pass over a 500 m × 1,000 m site; number of lines = 500/32.8 = 15.2 → 16 lines → 16 km of flying ≈ 3,200 s ≈ 53 min of line time, so roughly 4–5 battery sets of 20 minutes with turns and transit.
Expected accuracy: horizontal ≈ 1–2 × GSD = 2–4 cm, vertical ≈ 2–3 × GSD = 4–6 cm.

Example 5 — Accuracy check on independent checkpoints. Easting residuals (survey minus model) at five checkpoints: +0.021, −0.015, +0.030, −0.024, +0.008 m.

Σ(residual²) = 0.000441 + 0.000225 + 0.000900 + 0.000576 + 0.000064 = 0.002206 m².
RMSE = √(0.002206/5) = √0.0004412 = 0.021 m = 21 mm.
Against a specification of 2 × GSD = 40 mm for a 2 cm GSD survey, the block passes.

Method	Typical accuracy	Speed over 50 ha	Best suited to	Main limitation
Chain / levelling	mm–cm on points	Days	Small plots, verification	Very slow, few points
Total station	5–10 mm	2–4 days	Detailed topo, built-up sites	Line of sight, point-by-point
DGPS / RTK GNSS	10–20 mm	1–2 days	Control, open ground	Needs sky view; still point-wise
Drone photogrammetry	2–5 cm H, 4–8 cm V	1 day	Topo mapping, volumes, orthophoto	No canopy/underwater penetration
Drone / airborne LiDAR	3–10 cm	Hours–1 day	Vegetated corridors, forests	Costly hardware, heavier processing
Satellite stereo	0.5–5 m	Same day (archive)	Regional planning	Coarse resolution, cloud cover

Where it is used

Rural land records: SVAMITVA drone mapping of village *abadi* areas at about 5 cm GSD to issue property cards — executed by Survey of India with State revenue departments.
Highways and railways: DPR corridor surveys, LiDAR/photogrammetric DTMs for alignment and earthwork, and monthly drone orthomosaics for progress and quantity verification on NHAI and NHIDCL packages.
Mining and quarrying: monthly stockpile and pit volumetrics, lease boundary checks and mine-plan updates for IBM/State mining department compliance; typical volume accuracy of 1–3%.
Urban and water infrastructure: smart-city base maps, property tax surveys, canal and reservoir siltation studies, river morphology, flood mapping for State irrigation and disaster-management departments.
Construction and real estate: site topography before design, cut-and-fill balancing, 4D progress monitoring, as-built orthophotos; drone-plus-BIM workflows in large contractors (L&T, Tata Projects, AFCONS).
Job roles: survey engineer, drone pilot/remote pilot (RPC holder), photogrammetrist, GIS/geomatics analyst, UAV data processing engineer in firms such as Survey of India empanelled agencies, RITES, WAPCOS, AECOM, Jacobs and drone-service start-ups; also agriculture, insurance loss assessment and power-line inspection.

Interview questions

Q: Why is 60% forward overlap specified? A: So that every ground point appears on at least two consecutive photographs, which is the minimum needed for a stereoscopic model, with margin for drift, crab and height variation.
Q: What is relief displacement and is it a problem or a tool? A: The radial outward shift of images of points above datum, d = r·h/H. It distorts the photo as a map, but it lets you compute the height of an object from a single photograph.
Q: Difference between a photograph and a map? A: A photograph is a central projection with a varying scale and relief/tilt displacements; a map is an orthogonal projection at a uniform scale. Orthorectification converts one into the other.
Q: What is the difference between a DSM and a DTM? A: A DSM is the top surface including buildings and vegetation; a DTM is the filtered bare-earth surface used for contours, cross-sections and earthwork quantities.
Q: If a drone has RTK, do you still need GCPs? A: Fewer, but you must still place independent checkpoints to verify accuracy, and some GCPs are advisable to control vertical datum and prevent systematic doming.
Q: What is GSD and how do you decide flying height? A: GSD is the ground size of one pixel, GSD = pixel size × H/f; invert it to get the height needed for the specified GSD, then check it against the 120 m legal limit and obstacle clearances.
Q: What accuracy can you promise from a 2 cm GSD drone survey? A: Roughly 1–2 × GSD horizontally and 2–3 × GSD vertically with good control — about 2–4 cm and 4–6 cm — and it must be demonstrated on checkpoints, not assumed.
Q: Where will photogrammetry fail and what would you use instead? A: Under tree canopy, under water, and on featureless surfaces; use LiDAR for canopy, bathymetric methods for water, and total station/GNSS for isolated hidden points.
Q: What is the floating mark principle? A: Two half-marks in a stereoscopic viewing system fuse into one mark that appears to float; raising or lowering it until it sits on the terrain gives the parallax and hence the elevation of that point.
Q: What permissions do you need to fly a survey drone in India? A: Register the drone for a UIN on Digital Sky, fly with a Remote Pilot Certificate holder, check the airspace map — no prior permission below 120 m in a green zone, ATC/MoCA permission in yellow zones, and Government clearance in red zones. Rules are amended often, so verify before each job.
Q: What is bundle block adjustment? A: A simultaneous least-squares solution of the collinearity equations for all images, computing every photo's exterior orientation (and refining camera calibration) from tie points and ground control.
Q: Why do drone surveys use much higher overlaps than classical aerial photography? A: SfM matching relies on each point being seen in many images from different angles for robust tie points and self-calibration; small sensors and varying attitude also make low overlap risky.

Common mistakes

Using flying height above MSL instead of above the ground/datum of interest in the scale and relief-displacement formulas — h must be measured consistently.
Mixing units in d = r·h/H: r and d must be in the same photo units (mm), h and H in the same ground units (m).
Forgetting that photo scale varies across undulating terrain and quoting a single scale as exact.
In the number-of-photos calculation, applying overlap to only one direction, or omitting the two extra photographs at each strip end.
Confusing DSM with DTM, and then computing earthwork volumes that include trees and buildings.
Placing all GCPs in the middle or along one edge, producing doming, and reporting accuracy from the GCPs themselves instead of from independent checkpoints.
Assuming drone flying is unregulated: flying above 120 m, near an airport, or without a UIN and Remote Pilot Certificate is an offence under the Drone Rules.