corner detection
Enable an AI agent to recognise a corner-detection task, assess whether its detected image locations are useful and reliable, and decide how to configure, rerun or use the detector.
Bundle → Layer → Finding → Questions Filled
5 bundles · 9 layers · 14 findings · 22 questions
Corner meaning and task Establishes what counts as a corner and what the corner set must support.
Operational corner definition
Makes the detector's acceptance concept explicit.
Qualifying image evidence
Record the local evidence required for a corner and the treatment of straight edges, textured regions and junctions.
- What observable image pattern qualifies as a corner under this detector's criterion? definition
- How does the criterion distinguish a corner from a straight edge, blob or other interest point? boundary
Downstream corner requirements
Connects detection choices to the intended use of the locations.
Required corner behaviour
Record whether the task prioritises geometric vertices, repeatable interest points, positional accuracy or image coverage.
- Does the task require interpretable vertices or repeatable image locations regardless of physical meaning? boundary
- What corner count, coverage and localisation requirements must be met before the result can be used? action
Image evidence and support Captures how the input image and spatial neighbourhood make corner evidence available.
Image preparation
Records transformations that affect corner evidence and its interpretation.
Prepared image lineage
Identify the evaluated image representation and any conversion, smoothing, resizing or cropping applied before detection.
- Which source image and preprocessing operations produced the pixels evaluated by the detector? provenance
- How are locations in the prepared image mapped back to the source image? measurement
Neighbourhood and validity
Defines spatial support, detection scale and the region where evaluation is meaningful.
Corner support region
Record the neighbourhood or scale used to assess each candidate and the handling of borders and masked pixels.
- Over what neighbourhood and scale levels is corner evidence evaluated? measurement
- Are candidates near borders or masked regions excluded, padded or evaluated with incomplete support? boundary
Response, selection and localisation Explains how image evidence becomes a reproducible set of reported corner locations.
Detector response
Identifies the implemented criterion and the meaning of its output.
Response interpretation
Record the detector implementation, configuration and score semantics without assuming that response values are probabilities.
- Which algorithm, implementation version and parameter values generated the candidate responses? provenance
- What does a response value mean, and under which conditions can responses be compared? definition
Reported corner set
Defines candidate acceptance, duplicate handling and final position reporting.
Selection and coordinate contract
Record how thresholds, suppression, count limits and refinement produce the reported coordinates.
- Which threshold, non-maximum suppression, separation and count rules determine the retained candidates? action
- What coordinate convention and refinement procedure define each final location, and how are failed refinements handled? measurement
Corner quality and use decisions Evaluates whether the corner set remains accurate and useful under the intended operating conditions.
Evaluation protocol
Defines the reference evidence and measurements used to judge detection quality.
Repeatability and position evidence
Record evaluation images, reference locations or mappings, correspondence tolerances and treatment of non-overlapping regions.
- What reference annotations or known image transformations support evaluation, and where did they come from? provenance
- How are repeatability and localisation error calculated, including matching tolerance and the evaluated overlap? measurement
Failure diagnosis and next action
Connects observed corner-set limitations to justified operational decisions.
Corner-set acceptance
Record empty results, excessive clustering, unstable locations and runtime limits alongside evidence for accepting or revising the detection.
- Which observations distinguish insufficient corner evidence from unsuitable thresholds, scale, preprocessing or an execution failure? measurement
- Given the task's acceptance criteria, should the agent use the result, adjust the configuration, change the detector or request a different image? action
Evidence and external alignment What the world already says about this thing, gathered so the model can be checked against it.
Reported evidence
Findings from the breadth pass, kept separate from the structural claims.
Check these first
Recalled without web access and unsourced; every item is a lead to verify.
- Which of these check these first hold for the sense of corner detection this model covers, and on what evidence? provenance
Kinds and varieties
Recalled without web access and unsourced; every item is a lead to verify.
- Which of these kinds and varieties hold for the sense of corner detection this model covers, and on what evidence? provenance
Real-world use
Recalled without web access and unsourced; every item is a lead to verify.
- Which of these real-world use hold for the sense of corner detection this model covers, and on what evidence? provenance
Typical measurements
Recalled without web access and unsourced; every item is a lead to verify.
- Which of these typical measurements hold for the sense of corner detection this model covers, and on what evidence? provenance
Failure modes and hazards
Recalled without web access and unsourced; every item is a lead to verify.
- Which of these failure modes and hazards hold for the sense of corner detection this model covers, and on what evidence? provenance
Neighbouring kinds and how to tell them apart
Recalled without web access and unsourced; every item is a lead to verify.
- Which of these neighbouring kinds and how to tell them apart hold for the sense of corner detection this model covers, and on what evidence? provenance
Classifiers Filled
- Family
- Thing Registry
- Category
- Information and virtual systems
- Entry kind
- thing
- Plane
- INF
- Domain
- INF.MED
What it is Filled
Corner detection is a computer vision operation that locates image points with substantial local intensity variation in multiple directions, typically corresponding to corners, junctions or other distinctive points useful for matching and tracking.
In scope
- Operational definitions of image corners and corner-like interest points
- Input image preparation, coordinate conventions and valid detection regions
- Detector response, neighbourhood support and parameter choices
- Corner selection, duplicate suppression and localisation
- Evaluation of corner stability and suitability for downstream tasks
Out of scope
- General edge, line, blob and object detection
- Physical or architectural corners independently of their image evidence
- Feature descriptor construction and descriptor matching
- Tracking, optical flow and motion estimation beyond their requirements on detected corners
- Camera calibration and three-dimensional reconstruction
- Image acquisition hardware and general image restoration
Why it exists Filled
Enable an AI agent to recognise a corner-detection task, assess whether its detected image locations are useful and reliable, and decide how to configure, rerun or use the detector.
Distinguishing features Filled
- The operation selects image locations using an explicit corner criterion; detecting arbitrary salient locations alone does not establish corner detection.
- A candidate must satisfy the chosen criterion for locally multidirectional variation or boundary geometry, rather than merely lie on a strong straight edge.
- The primary result is a set of localised corner candidates, possibly accompanied by scores or scales, rather than a complete edge map or object segmentation.
- The operation can detect corners in one image without establishing correspondence to another image; correspondence belongs to matching or tracking.
- A detected image corner need not identify a physical vertex, and a physical vertex need not produce a detectable image corner.
What robots and AI may and may not do Filled
Must not
- Use corner features to identify or track people in images without consent or legal basis.
- Rely on corner detection alone for safety-critical decisions such as vehicle control without validation.
- Present feature matches as proof that two images show the same place or person.
Only with a human decision
- Deploying feature tracking in surveillance or safety-critical systems.
May
- Apply corner detection to images the agent is authorized to process.
- Explain common corner detection methods and their limits.
Moral aspects Filled
- Low-level vision tools become surveillance tools when used to track people.
- Errors in feature detection can propagate into unsafe machine behaviour.
Who is affected
- People in images
- Users of vision systems
Owners Filled
Steward
Nobody owns the technique; operators answer for each system that uses it.
Links to other meta-models Filled
parent
- Q114944645 - registry parent class
What else AI and robots need to interact with it Filled
Identity and identifiers required Filled
- Vercy registry: vr.tr.corner-detection
- Wikidata: Q697320 (https://www.wikidata.org/wiki/Q697320)
Direct properties not applicable Not applicable
- Operational corner criterion: Declared intensity-variation, local comparison, contour-geometry or learned criterion
- Input representation: Intensity image, multichannel image or image-derived contours; numeric range and conversion policy
- Detection support: Neighbourhood dimensions in pixels; scale levels or image-relative support where applicable
- Corner response semantics: Declared score definition, numeric range, ordering and normalisation; absent if unscored
- Acceptance and selection policy: Threshold rule, local-maximum rule, minimum separation, count limit and optional spatial quotas
- Localisation precision: Integer-pixel or subpixel coordinates, with coordinate origin and refinement method
- Valid detection region: Reference to the input image, region mask, excluded borders and coordinate transformation
- Detection outcome: Not evaluated, completed with corners, completed with no accepted corners, invalid input or failed
- Repeatability under declared changes: Matched-corner fraction or percentage, with correspondence tolerance and overlap rule
- Localisation error: Pixels or normalised image distance against declared reference locations
Plane INF: no invented physical properties.
Recognition not applicable Not applicable
Not applicable
Corner detection is a computational operation whose identity cannot be established from visual appearance alone.
Capabilities and actions required Filled
- Select and configure a corner criterion for the intended image conditions and downstream use.
- Detect and localise corners within a declared image region.
- Filter, rank and spatially distribute candidates using explicit selection rules.
- Compare configurations using localisation, repeatability, coverage and computational cost.
- Diagnose empty, clustered or unstable results and choose a justified adjustment.
- Export corners with coordinates and detection context for description, matching or tracking.
Hazards and failure modes required Filled
- Unstable detections can cause tracking failures or inaccurate motion estimates.
- Repeated textures can produce ambiguous correspondences and incorrect image alignment.
- Reliance on poorly validated detections in robot navigation can contribute to localization errors and collisions.
Standards and interfaces not applicable Not applicable
Not applicable
No single mandatory standard or official classification applies to corner detection as a generic image-processing operation.
Context of use required Filled
- Corner detection is used in computer vision to select distinctive image locations for matching, tracking and geometric analysis.
- It supports tasks such as image alignment, camera motion estimation and robot navigation.
- Its usefulness depends on whether detected points remain sufficiently stable across changes in viewpoint, lighting and image quality.
Sources Filled
- Wikidata item Q697320: corner detection - identity and sense of the item
- Wikipedia: Corner detection - general description of the item
Open questions
- Does the registry intend corner detection to cover both intensity-based interest-point detection and contour-based geometric corner detection?
- Which detector families and primary technical references should anchor the researched model's operational definitions?
- Should spatiotemporal or three-dimensional corner detection remain within this entry or link to neighbouring registered concepts?
- Which image transformations, datasets and localisation tolerances represent the intended Vercy use cases?
- Does an existing world model already cover this concept and therefore need to serve as its single publication?
Machine files
Provenance
thing registry research (pass 2) · unreviewed
Built from: models/things/publications/thing-q697320/spec.json