Skip to content

flight planning

Jump to bottom
Matthew R Aquin edited this page Mar 29, 2026 · 1 revision

Flight Planning

IronTrack's flight planning layer generates photogrammetric area surveys, corridor surveys, and LiDAR missions. It handles sensor geometry, flight line layout, procedure turns, and route optimization.

Overview

A flight plan starts with a boundary (bounding box or KML polygon) and a sensor definition, then moves through four stages:

  1. Sensor geometry — Compute GSD, swath width/height, and required AGL from the sensor optics and target resolution.
  2. Line generation — Lay out parallel flight lines across the survey area (grid) or along a centerline (corridor).
  3. Procedure turns — Connect consecutive flight lines with Dubins path turns and clothoid transitions.
  4. Route optimization — Reorder and optionally reverse flight lines to minimize total repositioning distance.

Terrain adjustment and trajectory smoothing (see Terrain & Trajectory) can be applied on top of any plan.

Sensor Geometry

The SensorParams struct holds the five intrinsic parameters of a frame camera: focal length, physical sensor dimensions (width × height in mm), and image dimensions (width × height in pixels).

Built-in Presets

Preset Focal Length Sensor Size Image Size
Phantom 4 Pro 8.8 mm 13.2 × 8.8 mm 5472 × 3648 px
Mavic 3 12.29 mm 17.3 × 13.0 mm 5280 × 3956 px
Phase One iXM-100 50.0 mm 53.4 × 40.0 mm 11664 × 8750 px

Custom sensors can be specified via CLI flags (--focal-mm, --sensor-width-mm, etc.).

Key Formulas

Ground Sample Distance (per axis):

gsd_x = AGL × sensor_width / (focal_length × image_width_px)
gsd_y = AGL × sensor_height / (focal_length × image_height_px)

The binding GSD is max(gsd_x, gsd_y) — the worst-case axis governs ASPRS compliance.

Required AGL for a target GSD:

AGL = target_gsd / max(sensor_width/(f × W_px), sensor_height/(f × H_px))

Swath dimensions:

swath_width  = AGL × sensor_width / focal_length    (cross-track)
swath_height = AGL × sensor_height / focal_length   (along-track)

Field of View:

FOV_horizontal = 2 × atan(sensor_width / (2 × focal_length))
FOV_vertical   = 2 × atan(sensor_height / (2 × focal_length))

Area Survey (Grid)

Grid-based flight line generation works in geodesic space (not UTM) to avoid zone-boundary issues over large survey areas.

Algorithm:

  1. Decompose the bounding box into cross-track and along-track components relative to the flight azimuth using Karney geodesic inverse.
  2. Compute the number of parallel lines: num_lines = ceil(cross_track_span / line_spacing) where line_spacing = swath_width × (1 − side_lap/100).
  3. For each line, compute the origin via a cross-track geodesic offset from the box centre, then step waypoints along the flight azimuth using Karney geodesic direct.
  4. Lines are generated in parallel via rayon. Each line is independent.

Waypoint spacing along each line is derived from the along-track photo interval: interval = swath_height × (1 − end_lap/100).

All waypoints are stored in radians internally. Conversion to degrees occurs only at export.

Corridor Survey

Corridor mode generates parallel offset passes along a polyline centerline (typically imported from KML). This is used for linear infrastructure: roads, pipelines, power lines, rivers.

Algorithm:

  1. Project the WGS84 centerline into UTM (single zone, determined from the centroid).
  2. Deduplicate consecutive vertices (< 1 µm tolerance).
  3. Compute offset distances for each pass:
    • 1 pass: centerline only.
    • 3 passes: −w/2, 0, +w/2.
    • 5 passes: −w/2, −w/4, 0, +w/4, +w/2.
  4. Generate offset polylines using miter joins (extend offset edges to intersection). When the miter spike exceeds 2× the buffer width, fall back to a bevel join (two points instead of one).
  5. Remove self-intersections via iterative segment intersection scanning.
  6. Resample each offset polyline at the requested waypoint interval.
  7. Convert back to WGS84 and pack into FlightLines.

LiDAR Missions

LiDAR planning replaces GSD-based geometry with point density kinematics. The primary quality metric is points per square metre rather than ground sample distance.

Key formulas:

point_density = PRR × (1 − loss) / (ground_speed × swath_width)
required_speed = PRR × (1 − loss) / (target_density × swath_width)
across_track_spacing = swath_width / (PRR / scan_rate)
along_track_spacing = ground_speed / (2 × scan_rate)

The along-track denominator uses 2 × scan_rate because oscillating mirrors scan bidirectionally. The loss fraction accounts for mirror deceleration and blind spots at scan edges.

Line spacing is driven by swath overlap percentage (typically 50%), similar to photogrammetric side-lap but computed from the LiDAR swath width.

Dubins Path Procedure Turns

Procedure turns connect consecutive flight lines with smooth, flyable curves. IronTrack evaluates all six canonical Dubins path types and selects the shortest:

Curve-Straight-Curve (CSC): LSL, RSR, LSR, RSL

Curve-Curve-Curve (CCC): LRL, RLR

The minimum turn radius is derived from physics, not user-configurable:

R_min = V² / (g × tan(bank_angle))

where V is true airspeed and bank_angle is the maximum allowed bank.

Clothoid Transitions

Raw Dubins paths have instantaneous curvature changes at the tangent points, which would require infinite roll rate. IronTrack inserts clothoid (Euler spiral) transitions that ramp curvature linearly:

  • Entry clothoid: Curvature ramps from 0 to 1/R over a distance proportional to the bank angle and roll rate limit.
  • Exit clothoid: Curvature ramps from 1/R back to 0.
L_clothoid = V × bank_angle / max_roll_rate

Clothoid length is capped at 40% of the arc length to prevent endpoint drift. Integration uses a midpoint-incremental method with a minimum of 4 steps per clothoid.

The roll rate limit prevents autopilot overshoot and gimbal destabilisation on camera platforms.

Route Optimization (TSP)

Flight line ordering is an Asymmetric Travelling Salesman Problem. IronTrack solves it with a two-phase heuristic:

Phase 1 — Nearest Neighbour:

Starting from the launch point, greedily visit the closest unvisited flight line. Each line can be flown in either direction; the algorithm picks the direction with the shortest entry distance.

Phase 2 — 2-opt Improvement:

Iteratively reverse segments of the tour when doing so reduces total repositioning distance. A swap is accepted if the gain exceeds 1×10⁻³ metres. The loop terminates when a full pass produces no improvement, or after 1,000 iterations.

All distances use the Karney geodesic (not Haversine), and total repositioning distance is accumulated with Kahan summation.

The result is a RouteResult containing the optimised line order, direction flags (forward/reversed), and total repositioning distance in metres.

Deep Dive: Implementation Notes

Terrain-Aware AGL Adjustment

When the --terrain flag is active, the flight line generator queries the Copernicus DEM at each waypoint and raises altitude where terrain would violate the side-lap or GSD target. This is a per-waypoint adjustment on top of the nominal AGL — it only raises, never lowers.

Kahan-Summed Geodesic Tracing

Flight line waypoints are placed by repeatedly calling geodesic_direct() with the same azimuth and step distance. Over a 100 km line, naive floating-point summation would accumulate centimetre-scale drift. Kahan compensated summation eliminates this.

ASPRS 2024 Compliance

IronTrack follows ASPRS Positional Accuracy Standards (2024 edition): RMSE-based accuracy classes with the 95% confidence multiplier eliminated. GSD heuristics follow the AT ≈ 0.5× pixel pitch rule. Minimum overlap is 60/30 (end/side), with 80/60 recommended.

Rolling Shutter Warning

For survey-grade PPK missions, a non-dismissable warning is displayed when the selected sensor uses a rolling shutter. Rolling shutter violates the collinearity equation assumption that all pixels in a frame share the same exterior orientation.

Forward Motion Compensation

smear_mm = V_ground × t_exposure × focal_length / AGL

If the computed smear exceeds the pixel pitch, the sensor will produce motion-blurred imagery. IronTrack warns the operator but does not enforce a speed limit (PIC retains authority).

Future: Energy-Weighted TSP (v0.5)

The current TSP uses geodesic distance as the cost metric. In v0.5, repositioning costs will incorporate wind triangle calculations, climb/descent energy, and battery state. The graph structure and 2-opt logic are designed to be reusable with energy-based weights.

Reference Documents

  • 03_photogrammetry_spec.md — GSD/FOV derivations, ASPRS standards.
  • 07_corridor_mapping.md — Polyline offsets, bank angles.
  • 13_lidar_survey_planning.md — Point density models, swath conflicts.
  • 39_flight_path_optimization.md — ATSP, NN+Or-opt, boustrophedon.
  • 50_dubins_mathematical_extension.md — Six Dubins types, clothoid, 3D airplane.
  • 36_asprs_2024_positional_accuracy.md — Accuracy class mapping.

Clone this wiki locally