Targets and perception

A target is a dynamic entity on the World: something the drones need to detect, recognize and keep observing. Each target has size, motion, appearance, reflectance, thermal signature, radar cross-section and a periodic sound source, and carries visual and acoustic sensitivity ranges: a drone that enters a sensitivity range may be "discovered" by the target. All perception criteria are computed on the server; the sensor views in the browser are visual approximations only.

Classes

Six class templates (packages/contracts/target/class_defaults.json) provide the default attributes; a new class only needs a new template entry. All values are simulated reference values.

Class

Size l × w × h (m)

Default motion

Default sensitivity ranges

Behavior after discovery

Person person

0.5 × 0.3 × 1.75

Random walk 1.4 m/s

Visual sector, half angle 60°, radius 300 m; acoustic hemisphere 300 m

Stops and looks for 10 s, then moves away at 2.5 m/s

Drone uav (third-party small multirotor)

0.35 × 0.35 × 0.10

Path or hover, 5 m/s

Visual cone, half angle 40°, 800 m

Moves away and returns home

Machine machine (generator and similar)

1.0 × 0.7 × 0.8

Stationary

None

—

Dog dog

0.9 × 0.3 × 0.6

Random walk 1.5 m/s, running 8 m/s

Visual sector, half angle 90°, 200 m; acoustic hemisphere 250 m

Barks and approaches to 20 m

Vehicle vehicle (passenger car)

4.5 × 1.8 × 1.5

Along a path, 10 m/s

Visual forward sector, half angle 60°, 150 m; acoustic hemisphere 100 m

Keeps driving

Object object (bag or box)

0.6 × 0.4 × 0.4

Stationary

None

—

A sensitivity range can be a sphere, hemisphere, cylinder or horizontal sector prism (directional), with a height interval, an azimuth offset relative to the heading and a back attenuation. They can be edited directly in the target panel, or set with "Adopt suggested radius", the value that the physical model suggests for the selected airframe.

Discovery criterion

Each target selects one criterion mode (field criterion_mode):

Mode

Criterion

Use

geometric (default)

The drone enters any enabled sensitivity range and stays for t_dwell

The range the visitor draws is exactly the boundary of discovery and of relay avoidance

physical (optional)

The drone is inside a sensitivity range and meets the visual criterion (line of sight, angular size, contrast and illuminance) or the acoustic criterion (margin of the arriving sound pressure per 1/3 octave band over the hearing threshold and background noise), staying for t_dwell

Study differences in airframe noise and visibility

The dwell time is t_dwell = 3·(1.2 − a) s, where a is the target's alertness (0.6 s at a = 1). Meeting the criterion emits the event target.discovered, which the interface shows as an alert with an acknowledge action. Background noise follows day and night profiles of the session's acoustic environment (rural, suburban, urban) and the simulation calendar, and rises with precipitation. Relay monitoring treats "zero discoveries" as a hard requirement.

Sensors

Sensor

Key facts

Zoom electro-optical eo_gx40

1/2.8 inch CMOS, native 3840 × 2160, focal length 4.8 to 48 mm (10× optical); analysis resolution 1080P by default

Night vision nir

Passive low-light imaging, plus active 850 nm illumination (built-in illuminator or the external long-range illuminator nir_ill_l)

Thermal lwir640

Uncooled 640 × 512, NETD 50 mK, lenses of 13, 25, 75 and 100 mm; white-hot and black-hot palettes

Millimeter-wave radar mmw77

77 GHz, reference range 180 m (RCS 1 m²), outputs tracks

Acoustic array mic8

8 microphones, outputs bearing and per-band signal-to-noise ratio

LiDAR mid360

Range depends on reflectance, cut off at 100 m

Detection, recognition and identification

Perception by imaging sensors unifies three levels through the Johnson criteria and the target transfer probability function (TTPF): the number of pixels across the critical dimension N = d_c·f / (R·p'), multiplied by factors for line of sight, contrast, illuminance and motion blur, gives the effective pixels N_eff.

Level

Meaning

Pixels for P ≥ 0.9

D detection

Something is there

3.5

R recognition

Class identified (person, dog, vehicle, drone, object)

14.0

I identification

Individual features identified

22.4

Definite capture: the same imaging sensor, for the same target, simultaneously has the target in its frustum, at least two thirds of the lines of sight visible, and N_eff ≥ N_req, held for 1.0 s of simulation time. By default N_req is the pixel count for P ≥ 0.9 at the required level (level_req of the target or task, R by default). The capture is lost after the conditions fail for 2 s. Radar, the acoustic array and LiDAR only produce detections and cueing (turning the gimbal towards the detection bearing) and cannot constitute a definite capture on their own.

The HUD of the FPV view shows the pixel progress towards the nearest target in the frustum (for example 9.6 / 14 px) and the current limiting factor (frustum, line of sight, pixels, illuminance, contrast or blur), which explains why a capture has not happened yet.

Perception quantification and field views

For each (target, drone) pair, the perception readings give the arriving sound pressure, threshold and margin, angular size, equivalent pixels, contrast, and per-modality and combined display probabilities of detection (GET /sessions/{sid}/perception/quant). The criteria themselves do not change; the display probability is calibrated so that the criterion boundary equals 0.5.

The field view in the viewport toolbar switches among none, acoustic, visual, thermal and radar: each drone shows a spherical scatter field decreasing with distance (five gray-level bands and iso-shells), overlaid with the threshold shell of the selected target, which shows where a drone would be heard or seen; the target's perception volume uses transparency for probability and point shape and gray level for the modality. Physical quantities are computed on the server as profile tables; the browser only samples them deterministically and draws.