GaitAI ForensicSearch
Search recorded CCTV by movement, timeline and event path.
ForensicSearch indexes movement and event evidence in recorded footage so authorized investigators can narrow hours of multi-camera video by time, path, movement pattern or tracked subject.
Problem
After an event, investigators may need to manually review large amounts of video across cameras.
What GaitAI does
Uploaded footage is indexed by movement: trajectories, events and appearance-agnostic movement descriptors become searchable, producing candidate segments and camera-wise trails instead of raw hours.
Who uses it
- Enterprise security teams
- Campus investigation units
- Transport-hub security
- Retail loss-prevention (authorized use)
What they receive
Example report values — illustrative, not measured performance claims.
Results are candidates for human confirmation — the system narrows where to look; the investigator decides what was found. All use sits inside the organisation's lawful-purpose and retention policies.
Why it matters
An investigation that would take a team days of manual scrubbing becomes a set of ranked candidate segments — reviewed, confirmed and documented by a human investigator.
Example workflow
- 01Authorized investigator uploads or selects footage
- 02Pipeline indexes movement and events
- 03Investigator queries by time, zone, path or pattern
- 04Candidate segments are ranked for review
- 05Confirmed findings compile into a review pack
Deployment
- Post-event, access-controlled workflow
- Case-scoped footage handling and retention
- Query audit logs per case
- Respects lawful purpose, access control and retention policy
Research basis
Published work is the research foundation for the capabilities this product is built on. That record covers the movement-analysis methods themselves — it is not a validation of this product's outputs, and there is no clinical validation study, accuracy benchmark or trial result behind them.
- Person re-identification
- Gait recognition & biometrics
- Movement biometrics
- Gait recognition & biometrics
- Advancements in Artificial Intelligence for Biometrics: A Deep Dive into Model-based Gait Recognition TechniquesEngineering Applications of Artificial Intelligence · 2024
- Journey into Gait Biometrics: Integrating Deep Learning for Enhanced Pattern RecognitionDigital Signal Processing · 2024
- Deep learning pipelines for recognition of gait biometrics with covariates: a comprehensive reviewArtificial Intelligence Review · 2023
- Data Preprocessing and Feature Selection Techniques in Gait Recognition: A Comparative Study of Machine Learning and Deep Learning ApproachesPattern Recognition Letters · 2023
- Intra-class variations with deep learning-based gait analysis: A comprehensive survey of covariates and methodsNeurocomputing · 2022
Evidence status
GaitAI Evidence Index
3 categories documented
What is documented for this module, and what is not. Research foundation is not product validation — the two are listed separately on purpose.
- Research foundationAvailable
- 5 published records directly address a capability this module is built on (Gait recognition & biometrics).
View evidence for Research foundation
Claim / capability: 5 published records directly address a capability this module is built on (Gait recognition & biometrics).
Status: Available
Applicability: The cited work is specifically about a method or movement signal this module draws on.
Limitation: Research relevance is not evidence that this product's outputs have been validated.
- publicationAdvancements in Artificial Intelligence for Biometrics: A Deep Dive into Model-based Gait Recognition Techniques · Engineering Applications of Artificial Intelligence, 2024
- publicationJourney into Gait Biometrics: Integrating Deep Learning for Enhanced Pattern Recognition · Digital Signal Processing, 2024
- publicationDeep learning pipelines for recognition of gait biometrics with covariates: a comprehensive review · Artificial Intelligence Review, 2023
- publicationData Preprocessing and Feature Selection Techniques in Gait Recognition: A Comparative Study of Machine Learning and Deep Learning Approaches · Pattern Recognition Letters, 2023
- publicationIntra-class variations with deep learning-based gait analysis: A comprehensive survey of covariates and methods · Neurocomputing, 2022
Inventory reviewed .
- Architectural research relationshipNot yet published
- No further shared-platform research relationship is recorded beyond the research foundation above.
View evidence for Architectural research relationship
Claim / capability: No further shared-platform research relationship is recorded beyond the research foundation above.
Status: Not yet published
Applicability: Shared capabilities such as pose estimation or temporal modelling are used by many modules.
Limitation: A shared-platform relationship is the weakest research link and must not be read as validating this module.
No supporting source is published for this status.
Inventory reviewed .
- Product specificationAvailable
- The module has a documented specification: inputs, pipeline, outputs and stated limitations.
View evidence for Product specification
Claim / capability: The module has a documented specification: inputs, pipeline, outputs and stated limitations.
Status: Available
Applicability: Documented inputs, intended workflow and output design.
Limitation: Documentation does not establish a released implementation or deployment readiness.
- specificationProduct documentation
Inventory reviewed .
- Product implementation evidenceNot yet published
- No module-specific release or deployment record is published here.
View evidence for Product implementation evidence
Claim / capability: No module-specific release or deployment record is published here.
Status: Not yet published
Applicability: Production implementation is distinct from this website's interactive prototypes.
Limitation: The browser pose explorer is not the full clinical or operational product pipeline.
- Context · not validation evidenceProduct and deployment context
Inventory reviewed .
- Interactive demoAvailable
- A sample output is available to explore. Illustrative demo — example values, not a measured result.
View evidence for Interactive demo
Claim / capability: A sample output is available to explore. Illustrative demo — example values, not a measured result.
Status: Available
Applicability: An interactive explanation of the intended output and workflow.
Limitation: Sample outputs contain illustrative values and are not measurements of a visitor or product performance.
- prototypeExplore illustrative output
Inventory reviewed .
- Benchmark validationNot yet published
- Validation pending publication. No product benchmark numbers are supplied.
View evidence for Benchmark validation
Claim / capability: Validation pending publication. No product benchmark numbers are supplied.
Status: Not yet published
Applicability: A benchmark needs a defined dataset, sample count, environment, hardware and evaluation method.
Limitation: Accuracy, sensitivity, specificity, latency and FPS cannot be inferred from a demonstration.
- Context · not validation evidenceResearch evidence inventory
Inventory reviewed .
- Product-specific validationNot yet published
- No study in the published record evaluates this module's output for a particular intended use.
View evidence for Product-specific validation
Claim / capability: No study in the published record evaluates this module's output for a particular intended use.
Status: Not yet published
Applicability: Validation must match the module, population and intended decision.
Limitation: A product description or related research paper does not establish validated performance.
- Context · not validation evidencePublished evidence inventory
Inventory reviewed .
- Pilot & deployment evidenceNot yet published
- No named pilot, deployment result or customer outcome is documented.
View evidence for Pilot & deployment evidence
Claim / capability: No named pilot, deployment result or customer outcome is documented.
Status: Not yet published
Applicability: Pilot and deployment evidence is assessed separately from research.
Limitation: Use-case journeys describe intended workflows, not customer results.
- Context · not validation evidenceTrust Center
Inventory reviewed .
- Product-specific clinical validationNot yet published
- No product-specific clinical validation study is published in this library.
View evidence for Product-specific clinical validation
Claim / capability: No product-specific clinical validation study is published in this library.
Status: Not yet published
Applicability: Clinical applicability depends on the intended use; safety modules are not presented as clinical tools.
Limitation: No output replaces clinician judgement or establishes a diagnosis.
- Context · not validation evidenceResponsible use and intended context
Inventory reviewed .
- Regulatory clearanceNot claimed
- No medical-device clearance, regulatory approval or compliance status is claimed.
View evidence for Regulatory clearance
Claim / capability: No medical-device clearance, regulatory approval or compliance status is claimed.
Status: Not claimed
Applicability: Regulatory applicability depends on the intended use and deployment jurisdiction.
Limitation: Research publications, patents and prototypes are not regulatory clearances.
- Context · not validation evidenceWhat GaitAI does not claim
Inventory reviewed .
Inventory reviewed How GaitAI states evidence →
Sample output
Privacy & responsible use
Designed to support consent-based capture, encrypted transfer, configurable retention and role-based access, with activity logging and skeleton-only processing modes. Deployment requirements depend on the applicable environment, organization and jurisdiction. Privacy-aware architecture does not automatically mean anonymity: pose, gait and trajectories may still carry identifying information. Movement analytics are designed to run without identifying individuals unless a deployment lawfully and explicitly requires it, and identity-related capabilities are intended only for lawful, authorized deployments with appropriate governance, access control and auditability. Outputs are decision support for trained operators — not autonomous enforcement. Watchlist, ReID, AccessMotion and ForensicSearch require documented lawful authority, consent where applicable, purpose-limited access, audit logs and human review. Candidate associations are not proof of identity and must not trigger automatic denial, enforcement or adverse decisions.
Context & limitations
- Candidates require human confirmation
- Index quality depends on footage quality
- Deployment must respect access control, lawful purpose and retention policy
Related SecureVision products
All products
Cross-camera movement correspondence, by gait.
Produces confidence-based candidate matches across multiple camera feeds using movement and body-level signatures rather than face-only matching. Candidates are for trained review — never proof of identity.
Anomalies surfaced — without identity matching as a prerequisite.
Surfaces candidate movement events — loitering, running, restricted-zone entry, tailgating-like patterns and perimeter events — for operator review, without requiring identity recognition.
Movement intelligence — with task-relevant movement representations.
The architectural privacy layer for GaitAI: designed to support skeleton-only analytics, face blur, role-based access, configurable retention and audit logs in privacy-sensitive environments.
Follow a person across live cameras, or search recorded footage? Compare ForensicSearch and ReID