NexuSec
Technical descriptions
The descriptions and static frames below explain the illustrated arrangements, including alternative views and selectable scenarios. Screenshots show the website examples; their accompanying text identifies what each example represents.
In these operational descriptions, visual observations mean information derived by applying machine learning to visual information: images, video frames, depth images, point clouds or inspection imagery. Models can detect and segment people and objects, estimate pose and placement, track movement, recognise actions, retrieve appearance candidates and identify anomalies. Sensor fusion associates selected visual outputs with other sensor measurements and operational records. Decorative drawings and website simulations remain illustrations rather than model outputs. Sensor-fusion methods and technology catalogue.
Investigation film
The film follows an investigation from a description back to supporting footage.
View the product example · Permanent figure link
The film presents the sequence of starting with a remembered description, searching selected video and returning to the relevant recorded moment for review. Machine learning applied to visual information supports appearance retrieval; sensor fusion can associate candidates with selected position, access and incident records. Source footage supplies context for the candidate result.
The film attempts to start with sound when it comes into view and pauses off-screen. If the browser blocks automatic playback, a Play with sound button starts it after a click. Manual playback and sound controls remain available, and a manual pause is retained. Reduced-motion preferences prevent automatic playback. This static frame is taken from the actual film at five seconds. The original video is separately accessible, while the written descriptions preserve the operational explanation independently of playback.
Person-appearance search
A written appearance description retrieves candidates for source-video review.
View the product example · Permanent figure link
Machine learning applied to visual information encodes clothing and carried appearance to retrieve candidate person appearances from a description in selected sources and dates. A reference image is another supported query route. Each candidate is associated with its camera and appearance time so the operator can open the original recording.
Sensor fusion can associate visual model outputs with selected spatial sensors, identifiers, timing, access or incident records, retaining source links and association uncertainty. The drawing illustrates the workflow and its evidence connections. Search candidates do not by themselves establish identity, and an action or audio alert requires assessment in context. The website visual is authored presentation content rather than a running search or model service.
Vehicle-appearance search
A vehicle description retrieves candidate vehicle appearances.
View the product example · Permanent figure link
Machine learning applied to visual information uses a delivery-van description to retrieve several candidate vehicle appearances. Source and date selections define the footage being searched. The candidate images are compared and the surrounding recording reviewed before a conclusion is drawn.
Sensor fusion can associate visual model outputs with selected spatial sensors, identifiers, timing, access or incident records, retaining source links and association uncertainty. The drawing illustrates the workflow and its evidence connections. Search candidates do not by themselves establish identity, and an action or audio alert requires assessment in context. The website visual is authored presentation content rather than a running search or model service.
Action and event analysis
Movement over multiple frames contributes to recognising an action or configured event.
View the product example · Permanent figure link
The illustration connects an original frame, successive observations and a timeline. Machine learning applied to visual information interprets motion and interactions across frames, such as running, climbing or aggressive interaction, rather than relying on one isolated still. Candidate events include timing, location, source camera and an event reason.
Sensor fusion can associate visual model outputs with selected spatial sensors, identifiers, timing, access or incident records, retaining source links and association uncertainty. The drawing illustrates the workflow and its evidence connections. Search candidates do not by themselves establish identity, and an action or audio alert requires assessment in context. The website visual is authored presentation content rather than a running search or model service.
Multimodal event context
Video, timing and available audio contribute to a reviewable event narrative.
View the product example · Permanent figure link
The illustration brings a scene image, time-indexed activity and sound annotations together. Machine learning applied to visual information derives scene, action and interaction features. Where audio is available, sensor fusion can combine these with audio analysis, timestamps, spatial measurements and selected operational records to support a readable narrative and questions about the event. Spoken notifications and possible distress alerts can be linked to the supporting footage for operator review.
Sensor fusion can associate visual model outputs with selected spatial sensors, identifiers, timing, access or incident records, retaining source links and association uncertainty. The drawing illustrates the workflow and its evidence connections. Search candidates do not by themselves establish identity, and an action or audio alert requires assessment in context. The website visual is authored presentation content rather than a running search or model service.
Model labelling, training and evaluation
Labelled data, training and evaluation support models for selected conditions.
View the product example · Permanent figure link
The workflow uses labelled visual information to train machine-learning models and evaluates their outputs against separate examples and relevant operating conditions. Camera views and scenarios are selected for the intended application; operator review of results informs further development.
Sensor fusion can associate visual model outputs with selected spatial sensors, identifiers, timing, access or incident records, retaining source links and association uncertainty. The drawing illustrates the workflow and its evidence connections. Search candidates do not by themselves establish identity, and an action or audio alert requires assessment in context. The website visual is authored presentation content rather than a running search or model service.
Visual machine learning, sensor fusion and screening technologies
The following describes possible arrangements across the illustrated product families. The catalogue is broad and non-exhaustive. Sources may be used individually or in selected combinations; not every technology is integrated, demonstrated or appropriate to every product. Specialist and research techniques are identified separately from routine checkpoint equipment.
From visual information to an associated event
Acquire live or recorded visual information and selected sensor or screening outputs. Retain source identifiers, timestamps, calibration, coordinate frames, quality and available uncertainty. Processing may occur at the sensor, on an edge device, on premises, in the cloud or across these locations.
Apply trained visual models to suitable frames, image sequences, depth data, point clouds or scan volumes. Examples include object detection, instance or semantic segmentation, keypoint and pose estimation, multi-object tracking, optical flow, activity recognition, visual-language models, image or text embeddings, appearance retrieval, OCR and anomaly detection. Models and training data are selected for the modality and task; an ordinary camera does not inspect concealed contents.
Align sources in time and, where applicable, transform them into a shared spatial representation. Associate observations using time, position, motion, identifiers, appearance and task context. Fusion may combine raw measurements, extracted features, tracks, model scores, decisions or workflow events, using rules, statistical estimation, probabilistic or Bayesian methods, state estimation, learned multimodal models or combinations.
For a lane example, visual machine learning identifies a passenger loading a tray and records subsequent interactions. Spatial measurements help follow the passenger and tray; RFID or a visible code can supply a tray identifier. Screening equipment contributes scan, alarm and completion records. Association connects these to the relevant passenger, tray, article and required task, retaining links to source video and images for the interaction flipbook.
A configured workflow can request correction, a recheck or staff review, and can communicate permitted hold or release requests. Missing, conflicting, delayed or low-quality evidence remains unresolved rather than being treated as verified completion. A visual action or equipment passage alone does not establish a successful screening result. Permissions, recipient verification, equipment interfaces and approved site procedures govern any handover or control.
Visual information and optical sensing
- Visible-light colour or monochrome images; CCTV, fixed, PTZ, panoramic, fisheye, line-scan, area-scan and high-speed cameras; still images and recorded or live video.
- Low-light and near-infrared imaging; short-, mid- and long-wave infrared imaging; thermal imaging and thermography; ultraviolet imaging; polarisation imaging; multispectral and hyperspectral images.
- Stereo and multi-view imaging; RGB-D cameras; structured light; time-of-flight depth cameras; event cameras; 3D surface images, photogrammetry and laser profilometry.
- Visual text and codes, including OCR, number-plate recognition, barcodes, QR codes and other machine-readable fiducials; image-derived object, interaction, placement and appearance features.
Range, position and movement
- Scanning, solid-state or flash LiDAR; laser rangefinders; 2D or 3D point clouds and range images.
- Radar, including Doppler, FMCW, ultra-wideband and millimetre-wave radar; ultrasonic range, proximity and occupancy sensors; active infrared distance sensors and light curtains.
- Ultra-wideband positioning, Bluetooth/BLE beacons and angle measurements, Wi-Fi positioning or channel-state sensing, GNSS/GPS and other satellite positioning, inertial measurement units, accelerometers, gyroscopes, magnetometers, odometry and encoders.
- Selected passive RF receivers, spectrum monitoring, radio direction finding and RF localisation can provide transmitter or device observations; these observations require separate association with a person, vehicle or incident.
Identifiers, access and custody context
- Passive or active RFID, including LF, HF, UHF and NFC arrangements; tag readers, transponders and supported real-time location tags.
- Tray, baggage, parcel, article, passenger-session and vehicle identifiers; boarding-pass or credential readers; access-control transactions, gate state, authorised work windows and escort records.
- Where selected and permitted, biometric matching outputs such as face, fingerprint, iris or palm/vein verification. Appearance similarity and tracking are distinct from verified personal identity.
- Linked interaction, custody, routing, storage, collection, delivery and recipient-verification records, including records from external logistics systems.
Contact, load and equipment sensing
- Load cells, scales, pressure mats, instrumented floors and seating, force or torque sensors, capacitive or inductive proximity sensors, contact switches and magnetic door contacts.
- Photoelectric break beams, passive infrared motion sensors, microwave occupancy sensors, conveyor encoders, motor current, belt speed, vibration, gate position and equipment interlocks.
- PLC, SCADA and building-management signals; equipment APIs, alarms, health, maintenance, availability and processing-state records; connected screening and staff-confirmation records.
Acoustic, fibre and perimeter sensing
- Microphones and acoustic arrays; ultrasonic or other acoustic measurements; vibration sensors, accelerometers, geophones and seismic ground sensors.
- Distributed acoustic, vibration, strain or temperature sensing on optical fibre; fibre Bragg grating sensors; fence-mounted vibration or microphonic cables; taut-wire and contact-based barriers.
- Buried pressure, seismic, magnetic or radio-frequency disturbance sensors; leaky-coaxial cable, electric-field and capacitance sensors; microwave barriers; active infrared beams and laser barriers.
- Environmental context such as temperature, humidity, light, precipitation, wind, air quality, smoke, particulates and gas measurements, used where relevant to activity or source quality.
Screening by X-ray and volumetric imaging
- Single-view or multi-view transmission X-ray; single-, dual- or multi-energy imaging; backscatter imaging; computed tomography (CT), cone-beam CT and other supported volumetric reconstructions.
- Spectral or photon-counting X-ray imaging; X-ray diffraction and coherent-scatter material analysis in specialist configurations; radiographic and CT inspection of baggage, parcels, cargo, vehicles, devices or components.
- Explosive detection systems (EDS), explosive detection systems for cabin baggage (EDSCB) and automated prohibited items detection systems (APIDS): equipment or software categories whose outputs may include images, detections, alarms, review decisions and task status.
- Machine learning on suitable X-ray images or CT volumes can locate and segment objects, classify supported categories, compare expected structure and flag candidates for review. Scan viewing, annotation and synthetic object insertion can also support model training and evaluation.
People, metal and concealed-item screening
- Walk-through, handheld and other metal detection equipment; electromagnetic induction and magnetometer-based detection; shoe metal detection and shoe screening arrangements.
- Active or passive millimetre-wave imaging and other selected body or concealed-item scanners; automatic anomaly or prohibited-item detection software; specialist terahertz imaging or spectroscopy where technically suitable.
- Ionising-radiation body imaging is a separate specialised modality, usable only where specifically authorised and appropriate. Listing it here does not propose adding it to an ordinary camera or AIVA portal.
Trace, vapour, liquid and chemical screening
- Explosive trace detection (ETD), including collected particles or swabs; explosive vapour detection (EVD); liquid explosive detection systems (LEDS), bottle and container screening; shoe explosive detection.
- Ion mobility spectrometry, differential mobility spectrometry, mass spectrometry and gas chromatography with suitable detectors, including GC/MS; chemical sensor arrays and electronic noses.
- Liquid or container analysis using selected optical, Raman, infrared, microwave, radio-frequency, dielectric, ultrasonic or magnetic-resonance methods, depending on equipment design and sample compatibility.
- Raman, surface-enhanced Raman and spatially offset Raman spectroscopy; infrared and Fourier-transform infrared (FTIR) spectroscopy; laser absorption, fluorescence, chemiluminescence, colorimetric and electrochemical sensing where appropriate to the substance and sample.
- Specialist or research arrangements can include terahertz spectroscopy, photoacoustic or photothermal analysis, nuclear quadrupole resonance and neutron-based material analysis. These require purpose-built equipment and task-specific validation.
Radiological and nuclear-material screening
- Passive gamma-ray and neutron detection; radiation portal monitors, personal radiation detectors, handheld survey instruments and radiation isotope identification devices.
- Scintillation or semiconductor spectroscopy and other appropriate detector technologies; source, energy-spectrum, alarm, location and review records supplied by the screening instrument.
Other specialist screening and detection inputs
- Chemical and biological monitoring can supply gas, vapour, aerosol or particle measurements, fluorescence or optical-scattering signals, immunoassay results, molecular-assay results or other instrument records. Sampling, laboratory confirmation and handling requirements depend on the target and method.
- Cargo or vehicle occupancy checks can contribute carbon-dioxide measurements, acoustic or vibration observations, heartbeat or motion sensing, thermal imagery and inspection results. These are candidate observations rather than automatic confirmation that a person is present.
- Document examination can contribute visible, ultraviolet or infrared images, optical security-feature checks, chip or credential validation and staff findings; these records can be associated with a session under the selected procedures.
Device, component and material inspection
- Visual or optical inspection, microscopy, borescopes, 3D surface measurement, industrial X-ray or gamma radiography, industrial CT and specialist neutron imaging.
- Ultrasonic testing, phased-array or time-of-flight ultrasonic techniques, acoustic emission, eddy-current and other electromagnetic testing, magnetic-particle inspection and liquid-penetrant testing.
- Thermography, shearography, strain or vibration analysis, electrical continuity, resistance, capacitance or impedance measurements and functional device tests, where appropriate to the inspection.
- Reference-image or reference-volume comparison, component recognition, missing or additional part detection, configuration comparison and anomaly analysis. These industrial inspection methods are distinct from routine passenger screening.
Operational records and human screening
- Search, pat-down, secondary inspection, canine screening, document examination, alarm resolution and staff-confirmed completion records. These contribute workflow evidence without being relabelled as electronic sensors.
- Queue, staffing, flight or transport schedules, resource availability, accessibility requirements, passenger instructions, authorised permissions and incident records can provide context for fusion, validation, forecasting and response.
- Any selected combination may use instrument-native analysis, separate machine-learning models, human review or a combination. Availability, compatibility and required approvals are determined for the actual equipment and installation.
Background references
These public references explain selected technology families. They do not certify NWPRO products or establish that any listed integration has been implemented.
- NIST: sensing and perception. Camera calibration and 3D sensing.
- NIST: quantitative information from images. Spatial, spectral, temporal and polarisation information.
- NIST: audiovisual sensor fusion. An example of combining information from different modalities.
- NPSA: perimeter intrusion detection. Perimeter sensor families and site context.
- ECAC: security equipment categories. Screening equipment categories; evaluation applies to specific configurations.
- DHS: baggage, cargo and people screening. Screening research and development.
- DHS: explosives trace detectors. Trace detection and ion mobility spectrometry.
- DHS: chemical analysis technologies. IMS, Raman and FTIR analysis.
- DHS: portable GC/MS. Field chemical analysis.
- NIST: microwave liquid analysis. An example of non-optical liquid identification.
- NIST: biological detection equipment. Biological detection and identification methods.
- GAO: air-cargo screening methods. Historical discussion of cargo screening and occupancy checks, not a current equipment approval list.
- IAEA: industrial non-destructive testing. Optical, radiographic, ultrasonic, electromagnetic and other NDT methods.
- PNNL: infrared and terahertz research. Specialist spectroscopic measurements.
- Nuclear quadrupole resonance research. An experimental material-detection approach.
- DHS: radiological detection instruments. Portal monitors, isotope identification and personal detectors.