Spatial and Kinematic Descriptors

Spatial and Kinematic Descriptors are tools used to analyze movement and position, offering insights into dynamic behavior through measurable parameters.

Spatial and Kinematic Descriptors explicitly characterize the spatial configuration and motion of declared points, landmarks, segments, rigid bodies, trajectories, poses, tracked objects, or other geometric entities represented in Behavioral Signal evidence. Spatial descriptors characterize where entities are located and how they are configured relative to declared coordinate frames or other entities. Kinematic descriptors characterize how position and orientation change over time through quantities such as displacement, distance, velocity, acceleration, angular change, and path geometry. Kinematics alone does not describe forces, torques, energy expenditure, causal mechanisms, or behavioral meaning. Essential semantic components of these descriptors include coordinate frame, dimensionality, units, calibration, entity identity, timing, and derivation method.

Meaning and Geometric Input Semantics

A Spatial or Kinematic Descriptor is a reproducible characterization of position, orientation, displacement, distance, spatial extent, configuration, trajectory geometry, linear motion, angular motion, or related geometric change under a declared coordinate system and temporal support. Descriptor inputs may come from directly instrumented coordinates, reconstructed markers, pose-estimation landmarks, inertial orientation sensors, model coordinates, tracked objects, center-of-mass estimates, or derived geometric signals. These different origins carry varying uncertainty and evidential status, affecting descriptor interpretation.

Geometric entities and representations include scalar coordinates, points, vectors, landmark sets, body segments, rigid-body poses, orientations, skeletons, polylines or trajectories, spatial regions, joint configurations, and object tracks. Storing several numbers together does not establish whether they represent coordinates, a vector, an orientation parameterization, or a model state; the geometric object and its coordinate frame must be declared explicitly.

Geometric Object	Minimum Identity Needed	Representative Descriptor	Primary Interpretation Risk
Point	Unique label, frame, and coordinates	Position vector	Confusing coordinate vector with direction or displacement
Vector	Origin and direction in frame	Displacement or velocity vector	Treating as position rather than relative difference
Landmark Set	Ordered set of points with identity	Multi-point configuration	Permutation of points alters meaning
Body Segment	Defined endpoints or landmarks, frame	Segment length, direction	Ambiguity of segment definition or attachment
Rigid-Body Pose	Position + orientation, frame	Pose vector + quaternion or matrix	Mixing orientation representation conventions
Orientation	Rotation relative to reference frame	Euler angles, quaternions	Singularities, ambiguities in angle parameterizations
Skeleton	Set of joints and segments, model definition	Joint positions and angles	Model assumptions affect meaning
Trajectory	Ordered sequence of positions over time	Path coordinates, velocity profile	Sampling density and smoothing affect geometry
Spatial Region	Defined boundary or volume in frame	Convex hull, bounding box	Defining region limits and their physical relevance
Joint Configuration	Set of joint angles or coordinates	Joint angle vector	Dependence on model and conventions

Spatial and Kinematic Descriptors differ fundamentally from waveform morphology, generic temporal descriptors, cross-signal dependence, and kinetics/dynamics. For example, a speed profile over time can exhibit waveform morphology, but the underlying trajectory describes spatial geometry. Event timing can indicate when motion occurs, while synchrony compares timing between motion signals. Forces or torques require additional dynamic evidence or modeling beyond kinematics. These concepts are related but not interchangeable.

Geometric-input provenance and reconstruction dependence are critical: coordinates can be measured directly, triangulated, inferred from a model, estimated by computer vision algorithms, integrated from inertial sensors, or transformed from another frame. A reconstructed landmark or orientation remains derived evidence whose calibration, model assumptions, occlusion handling, filtering, and uncertainty influence every downstream spatial or kinematic descriptor.

Coordinate Frames, Dimensionality, and Spatial Calibration

Coordinate frames are defined by an origin, a set of basis axes with specified handedness and orientation, units of measure, and frame identity. Common frames include image/pixel frames, camera frames, laboratory/world frames, body-centered frames, segment-local frames, object-centered frames, and other declared frames. The same physical point will have different coordinate values in each frame; identical numeric coordinates in different frames do not imply identical physical locations.

Mappings between coordinate descriptions include translations, rotations, rigid transformations, scale changes, reflections, and projections. Such transformations require explicit specification of direction, source frame, destination frame, units, calibration or version, and time dependence when the frame moves. A descriptor invariant to one transformation may not be invariant to another. Transforming coordinates can change component-wise descriptors while preserving some geometric relations.

Spatial dimensionality includes 1D, 2D image-plane or planar geometry, projected or partial-depth representations, and full 3D geometry. Projection shortens distances, alters angles, hides depth motion, merges distinct 3D configurations, and changes apparent velocity. Interpreting 2D pose or displacement as 3D physical geometry requires justified reconstruction or planar-motion assumptions.

Calibration and spatial scale vary: pixel coordinates, normalized image coordinates, metric world coordinates, body-size-normalized coordinates, and model coordinates have different units and comparability. Camera intrinsics/extrinsics, marker scaling, model anthropometry, sensor alignment, lens distortion correction, and other calibration steps materially alter descriptor meaning.

Frame	Reference	Typical Units	Comparability or Invariance Risk
Image/Pixel	Camera sensor image plane	pixels	Varies with resolution, zoom, camera intrinsics; not metric
Camera	Camera center and optical axes	pixels, normalized units	Projection distortion; camera motion affects values
World/Laboratory	Fixed environment coordinate system	meters, centimeters	Requires calibration; may move relative to actor or objects
Body-Centered	Anatomical landmark or body origin	normalized length, meters	Depends on model accuracy; changes with pose
Segment-Local	Local segment coordinate system	normalized length, radians	Dependent on segment definition; may rotate with segment motion
Object-Centered	Object reference frame	meters, normalized units	Object pose changes affect values
Normalized	Scaled by body size or workspace	unitless ratios	Sensitive to denominator accuracy; loses absolute scale

Position, Displacement, Distance, and Spatial Extent

Position descriptors are component coordinates or frame-dependent location summaries of a declared entity. Absolute position specifies location in a coordinate frame, while relative position expresses location relative to another point, body, region, landmark, or origin. Position descriptors inherit coordinate-frame translation and orientation unless the definition explicitly removes these dependencies.

\begin{matrix} Δ r_{a \to b} = r (t_{b}) - r (t_{a}) \\ D_{a \to b} = \sqrt{Δ r_{a \to b} \cdot Δ r_{a \to b}} \end{matrix}

Here, $r (t)$ is the declared position vector of one entity in one coordinate frame; $t_{a}$ and $t_{b}$ are ordered coordinates or times with $t_{b} > t_{a}$ ; $Δ_{r_{a → b}}$ is the vector displacement from the first to the second position; $·$ is the Euclidean inner product in that coordinate space; and $D_{a → b}$ is the displacement magnitude. Displacement magnitude is not path length, and Euclidean distance is valid only when the coordinate geometry and units justify that metric.

Position difference, displacement vector, displacement magnitude, pairwise distance, signed axis displacement, radial distance from a reference, and distance to a region or surface differ conceptually. A scalar distance removes direction, while signed component displacement depends on axis orientation.

Spatial extent and dispersion descriptors characterize landmark sets or tracked entities, including bounding dimensions, convex-hull-like extent where justified, radius from a center, spread of points, body or object span, occupied region, and pairwise-distance summaries. Geometric extent differs from temporal occupancy and from statistical dispersion of an unrelated scalar signal.

Normalization by body size, segment length, image dimensions, workspace size, or another reference is valid only when the denominator has scientific meaning. Normalization can improve some cross-participant or cross-camera comparability but removes absolute scale. Ratios become unstable when the reference dimension is poorly estimated or near zero.

Descriptor	Frame Dependence	Direction Preserved?	Common Misinterpretation
Absolute Position	Yes	Yes	Treated as relative or invariant
Relative Position	Yes	Yes	Confused with absolute position
Displacement Vector	Yes	Yes	Mistaken for speed or velocity
Displacement Magnitude	Yes	No	Confused with path length
Pairwise Distance	Yes	No	Interpreted as vector difference
Radial Distance	Yes	No	Assumed directional without reference
Spatial Extent	Yes	No	Conflated with temporal occupancy
Normalized Distance	Yes (scaled)	No	Treated as absolute scale

Orientation, Angles, and Configuration

Orientation describes the rotational configuration of a directed axis, segment, rigid body, or local frame relative to another declared frame. Orientation differs from position and can be expressed as absolute orientation in a fixed frame or as relative orientation between bodies or segments. Representations include angles, rotation matrices, quaternions, axis-angle, direction vectors, and other parameterizations.

Orientation parameterizations differ and are not equivalent at the coordinate level. Euler/Cardan angles depend on axis sequence and can encounter singular configurations (gimbal lock). Quaternions have sign ambiguity because opposite signs represent the same physical rotation. Rotation matrices contain redundant constrained elements, and axis-angle representations require angular conventions. Raw parameter components across different conventions should not be compared as if they represent the same physical descriptor.

θ = \arccos (\frac{u \cdot v}{‖ u ‖ ‖ v ‖})

Here, $u$ and $v$ are two valid nonzero vectors expressed in a compatible Euclidean coordinate frame; $·$ is their inner product; $‖·‖$ denotes Euclidean norm; and $θ$ is their unsigned principal angle. This relation does not encode rotation direction, joint-axis convention, or anatomical flexion/extension semantics and should not be used as a complete joint-angle definition when those conventions matter.

Joint-angle and segment-angle descriptors depend on model and convention. Joint angles depend on anatomical landmarks, local segment frames, joint coordinate systems, axis sequence, model constraints, and calibration pose. A model generalized coordinate differs from the simple geometric angle between two segment vectors unless explicitly defined equivalently.

Posture and configuration descriptors combine several positions or orientations, including inter-joint angles, relative segment orientations, normalized landmark configurations, center-relative coordinates, posture span, symmetry relations, and selected shape ratios. Entity labels and anatomical or object roles must be preserved; permuting landmark identities can describe a different configuration even if the point cloud geometry is unchanged.

Object	Reference Needed	Strength	Principal Ambiguity or Limitation
Direction Vector	Frame origin and axes	Simple, intuitive	Does not encode rotation magnitude
Absolute Segment Orientation	Fixed coordinate frame	Complete orientation	Sensitive to frame choice and calibration
Relative Orientation	Pair of segments or frames	Describes relative pose	Depends on reference conventions
Unsigned Vector Angle	Compatible vectors	Simple angle measure	No rotation direction or joint semantics
Euler/Cardan Angles	Axis sequence and frame	Compact angle description	Singularities, sequence-dependence
Quaternion	None (unit norm constraint)	Smooth rotation interpolation	Sign ambiguity
Rotation Matrix	Orthogonal matrix frame	Full rotation with constraints	Redundancy and numerical drift
Joint Coordinate	Model and anatomical definition	Biomechanical interpretation	Model-dependent, may differ from geometric angle

Linear Kinematic Descriptors

Finite-difference velocity and acceleration estimators are

\begin{matrix} v_{n} = \frac{r_{n} - r_{n - 1}}{t_{n} - t_{n - 1}} \\ a_{n} = \frac{v_{n} - v_{n - 1}}{t_{n} - t_{n - 1}} \end{matrix}

where $n$ is an ordered observation index; $r_n$ and $r_{n-1}$ are compatible position vectors; $t_n$ and $t_{n-1}$ are valid temporal coordinates with positive elapsed time; $v_n$ is the backward finite-difference velocity estimate; and $a_n$ is the finite-difference acceleration estimate from successive velocity estimates. Alternative central, forward, locally fitted, filtered, or model-based derivatives define different estimators.

Velocity vector, speed, component velocity, radial or tangential velocity relative to a declared reference, and path-aligned velocity differ conceptually. Speed is the nonnegative magnitude of velocity and discards direction; component velocities depend on frame axes; radial/tangential decompositions depend on the chosen center or path geometry.

Acceleration vector, acceleration magnitude, component acceleration, tangential acceleration, and normal/centripetal-like geometric components require defined trajectory and derivative estimates. Acceleration characterizes the rate of change of velocity and is not force unless mass and relevant dynamical models and external/internal forces are justified.

Derivative sensitivity is critical: velocity and especially acceleration amplify position noise, calibration jitter, tracking swaps, interpolation artifacts, and timing error. Smoothing before differentiation, differentiating before smoothing, polynomial or local fitting, and model-constrained estimation produce different kinematic descriptors and timing. The estimator type and its causal or future-inclusive status should be preserved.

Jerk and higher-order translational derivatives are used only when scientifically needed and when estimator fidelity supports them. Units and derivative conventions must be stated. Sensitivity to noise and preprocessing rapidly increases with derivative order. Higher-order derivatives should not be automatically interpreted as smoothness, motor quality, effort, comfort, or control strategy without separate behavioral or biomechanical arguments.

Angular Kinematic Descriptors

Angular displacement and orientation change are differences between rotations rather than ordinary subtraction of arbitrary orientation parameters. Relative rotation should be computed under the declared orientation representation and frame convention. Euler-angle differences can fail to represent the physically shortest or correct relative rotation when wrapping, axis sequence, or singularities matter.

Angular velocity and angular speed are time derivatives or finite changes of orientation under a valid rotational representation. Angular velocity is a vector quantity expressed in frames such as world, body, or sensor frames. Angular speed is a scalar magnitude. Units such as radians per second or degrees per second must be explicit.

Angular acceleration and higher angular derivatives are derivatives of angular velocity under a declared frame and estimator. Filtering, orientation unwrapping, quaternion sign continuity, sampling rate, sensor fusion, and frame transformations strongly affect these descriptors. Angular acceleration is not torque without an inertial/dynamical model and force/moment information.

Descriptor	Input Primitive	Units	Major Estimation or Interpretation Risk
Linear Velocity	Position vectors	m/s, units/s	Noise amplification, sampling rate
Speed	Velocity vector magnitude	m/s, units/s	Confusion with vector velocity direction
Linear Acceleration	Velocity vectors	m/s², units/s²	Highly sensitive to noise and filtering
Jerk	Acceleration vectors	m/s³, units/s³	Extreme noise sensitivity, interpretation ambiguity
Angular Displacement	Orientation parameters	radians, degrees	Representation-dependent computation
Angular Velocity	Orientation change rates	rad/s, deg/s	Frame dependence, unwrap errors
Angular Speed	Magnitude of angular velocity	rad/s, deg/s	Loss of directional information
Angular Acceleration	Angular velocity rates	rad/s², deg/s²	Sensitive to noise, requires consistent orientation

Trajectory Geometry and Path Descriptors

Discrete path length and straightness are defined as

\begin{matrix} L = \sum_{n = 1}^{N − 1} ‖ r_{n} - r_{n − 1} ‖ \\ S = \frac{‖ r_{N − 1} - r_{0} ‖}{L} \end{matrix}

where $N$ is the number of ordered valid trajectory samples with $N ≥ 2$ , $n$ is the trajectory sample index, $r_n$ is the position vector at sample $n$ in one compatible Euclidean frame, $L$ is the discrete path length, and $S$ is the endpoint-displacement-to-path-length straightness when $L > 0$ . The ratio $S$ lies between zero and one under these assumptions; a stationary trajectory makes the ratio undefined rather than automatically one. Path length depends on sampling density, smoothing, and spatial noise.

Trajectory descriptors include net displacement, path length, straightness or directness, tortuosity-like measures, cumulative turning, turning angles, curvature-like path descriptors, spatial envelope, excursion range, and residence in spatial regions. The geometric metric, sampling, coordinate frame, smoothing, and path parameterization must be declared where they affect the result.

Path geometry must be distinguished from timing along the path. Two motions can follow the same geometric path with different speed profiles, pauses, or directional timing. Conversely, similar speed distributions can occur on different paths. Descriptors must preserve whether they characterize geometry only, kinematic traversal, or combined space-time properties.

Spatial occupancy, region visitation, boundary approach, path-to-target distance, workspace coverage, and center-of-motion descriptors require a meaningful declared region or target geometry. Geometric occupancy of space differs from temporal occupancy duration unless time weighting is explicitly included. Region definition and coordinate frame must be preserved.

Pose, Relative Geometry, Uncertainty, and Provenance

Pose and skeleton descriptors derive from joint positions, segment lengths, joint configurations, body-centered coordinates, relative orientations, symmetry measures, center-of-mass or center-of-geometry estimates, posture extent, and selected inter-landmark relations. Anatomical structural connectivity differs from measured kinematic relations; a geometrically plausible pose does not imply anatomical correctness or behavioral state.

Relative and multi-entity geometric descriptors include inter-person distance, actor–object distance, relative bearing, facing orientation, overlap or proximity, hand-to-object distance, joint-to-target distance, or instantaneous spatial configuration. These remain geometric relations at declared instants or supports and must not be interpreted as cross-signal synchrony, social coordination, interaction intent, or causal influence without additional evidence.

Worked Example and Uncertainty/Provenance Audit

Consider a tracked upper-limb movement observed initially in camera coordinates and transformed to a calibrated world frame and subsequently to a body-centered frame.

Position: The tracked wrist point has camera-frame pixel coordinates converted to metric world coordinates via camera intrinsic and extrinsic calibration.
Relative Displacement: Displacement vectors between consecutive wrist positions are computed in the world frame.
Path Length vs. Net Displacement: Path length is estimated as the sum of Euclidean distances between consecutive points, while net displacement is the vector difference between initial and final positions.
Speed/Velocity: Finite-difference velocity is computed using backward differences in world coordinates; speed is the magnitude of velocity.
Acceleration: Derived via finite differences of velocity; sensitive to smoothing choices.
Orientation Descriptor: The forearm segment orientation is computed as a quaternion relative to the world frame.
Angular Velocity: Estimated as finite differences of orientation over time.
Pose/Actor–Object Relation: The distance between the hand and a target object is computed in the body-centered frame.
2D Projection: A projected 2D view (e.g., image plane) shows lost depth, obscuring true spatial relations.
Tracking Gap / Interpolation: Missing data segments are interpolated, affecting velocity and acceleration estimates.
Smoothing Choice: Applying a low-pass filter alters acceleration magnitude and path length estimates.

Throughout, descriptor definitions, entity and landmark identities, dimensionality, source and reconstruction methods, coordinate frames and transformations, handedness, calibration and scale, units, timestamps and sampling, support and validity masks, pose/model version, orientation representation and axis conventions, derivative estimators, filtering/interpolation, normalization, region/target geometry, uncertainty, implementation/version, and sensitivity findings are preserved.

This example illustrates that reproducible coordinate arithmetic and kinematic computation alone do not establish physical measurement accuracy, kinetics, causal mechanism, or behavioral interpretation.