armarx::fake_object_detector Namespace Reference

Geometric primitives for the fake object detector. More...

Classes
struct	Box
	An oriented bounding box in the global frame. More...
struct	Frustum
	A symmetric view frustum. More...

Enumerations
enum class	DetectionMode { Always , FieldOfView , LineOfSight }
	How strictly an object must be perceivable before it is reported. More...

Functions
Eigen::Vector3f	faceCenter (const Box &box, int faceIndex)
	Centre of faceIndex in global coordinates.
Eigen::Vector3f	faceNormal (const Box &box, int faceIndex)
	Outward unit normal of faceIndex in global coordinates.
int	frontFaceIndex (const Frustum &frustum, const Box &box)
	The face of box that points most directly at the camera - the "front facing side" of the bounding box.
std::optional< float >	intersectRayBox (const Eigen::Vector3f &origin, const Eigen::Vector3f &unitDirection, const Box &box)
	Intersect a ray with an oriented box using the slab method.
bool	isInFieldOfView (const Frustum &frustum, const Box &box, bool requireFullyInside)
	Whether box is inside the camera frustum.
std::vector< Eigen::Vector3f >	sampleFace (const Box &box, int faceIndex, int gridN)
	A gridN x gridN grid of sample points spread over faceIndex, in global coordinates.
float	visibleFraction (const Frustum &frustum, const Box &target, const std::vector< Box > &occluders, int gridN, float epsilon)
	The fraction of sample points on the front face of target that the camera can see.

Detailed Description

Geometric primitives for the fake object detector.

This header is deliberately free of Ice, ArmarX component and Simox types so that the visibility logic can be unit tested in isolation.

All lengths are in millimetres and all poses are in the global frame, matching the ArmarX/Simox convention.

Enumeration Type Documentation

DetectionMode

enum class DetectionMode

strong

How strictly an object must be perceivable before it is reported.

Enumerator
Always	Report every object, regardless of where the camera looks.
FieldOfView	Report an object once its bounding box enters the camera frustum.
LineOfSight	Additionally require a clear line of sight to the front face of the bounding box.

Function Documentation

faceCenter()

Eigen::Vector3f faceCenter	(	const Box &	box,
		int	faceIndex )

Centre of faceIndex in global coordinates.

faceNormal()

Eigen::Vector3f faceNormal	(	const Box &	box,
		int	faceIndex )

Outward unit normal of faceIndex in global coordinates.

frontFaceIndex()

int frontFaceIndex	(	const Frustum &	frustum,
		const Box &	box )

The face of box that points most directly at the camera - the "front facing side" of the bounding box.

Returns

A face index in [0, 6): 0 = +X, 1 = -X, 2 = +Y, 3 = -Y, 4 = +Z, 5 = -Z.

intersectRayBox()

std::optional< float > intersectRayBox	(	const Eigen::Vector3f &	origin,
		const Eigen::Vector3f &	unitDirection,
		const Box &	box )

Intersect a ray with an oriented box using the slab method.

Parameters

unitDirection

Must be normalized.

Returns

Distance from origin to the nearest intersection, or std::nullopt if the ray misses. Returns 0 when origin lies inside the box.

isInFieldOfView()

bool isInFieldOfView	(	const Frustum &	frustum,
		const Box &	box,
		bool	requireFullyInside )

Whether box is inside the camera frustum.

Parameters

requireFullyInside

If true, all eight corners must be inside. Otherwise it is enough that any corner or the box centre is inside.

sampleFace()

std::vector< Eigen::Vector3f > sampleFace	(	const Box &	box,
		int	faceIndex,
		int	gridN )

A gridN x gridN grid of sample points spread over faceIndex, in global coordinates.

Samples sit at cell centres, so they never lie exactly on an edge shared with an adjacent object. gridN == 1 yields the face centre alone.

visibleFraction()

float visibleFraction	(	const Frustum &	frustum,
		const Box &	target,
		const std::vector< Box > &	occluders,
		int	gridN,
		float	epsilon )

The fraction of sample points on the front face of target that the camera can see.

A sample is blocked when any occluder is hit closer than the sample itself. An occluder containing the camera (such as the head link the camera is mounted in) never blocks.

Parameters

occluders	Must NOT contain target; otherwise every ray is blocked by the target's own front face.
epsilon	Tolerance [mm] to absorb the hit exactly on the target's surface.

Returns

A value in [0, 1]; 0 if the face has no samples.