AI Vision Beyond Security CCTV

Cameras That Inspect, Measure and Count.

AiVisionBot is AI computer vision software by VYROX that puts cameras to work on the jobs that make money, not just the ones that watch doors: it finds defects on the line, grades produce, counts trucks and livestock, measures stockpiles and checks PPE, then raises an alert or switches a relay through IoTServa in seconds.

Explore 30 Use Cases Book a Site Assessment

30High-value use cases

12Core vision capabilities

1Engine, edge or cloud

Quality engineer with a tablet beside a production line watched by an overhead inspection camera

Defect Found, Line 2Surface scratch on unit 4,812. Reject gate triggered, photo logged to the QC report.

Works With IP and Industrial Cameras Edge Box or Cloud Inference Alerts Drive IoTServa Automations Video Can Stay on Site

Use Cases

30 Jobs Your Cameras Can Do, Ranked by Value.

Ranked by measurable customer value, commercial demand, technical maturity and how much of the vision engine they share. Pick your industry to see what fits.

#01 · IndustrialPilot

Manufacturing Defect Inspection

Scratches, weld faults, missing parts and surface defects caught on the line.

Fewer escapes and rework, measurable in weeks.

#02 · IndustrialPilot

Rubber Glove Inspection

Pinholes, tears, stains and thin spots spotted on every former at line speed.

Fewer rejected batches and customer claims.

#03 · Construction and MiningLive

Mining, Quarry and Heavy Plant

Truck trips, loading, payload fullness and plant movement tracked.

Every load counted, underloading flagged per trip.

#04 · Construction and MiningPilot

Stockpile Volume Measurement

Aggregate and material pile volumes from drone or fixed cameras.

Monthly stock-take without surveyors.

#05 · IndustrialPilot

Workplace Safety and PPE

Helmets, vests, restricted machinery zones and forklift proximity.

Fewer incidents and audit-ready proof of compliance.

#06 · Agriculture and EnergyRoadmap

Palm Fruit Bunch Grading

Ripe, unripe, overripe and empty bunches graded at the mill ramp.

Fair grading and higher oil extraction rate.

#07 · Transport and LogisticsRoadmap

Warehouse and Logistics

Parcel counting, dimensioning and damage spotted on the conveyor.

Fewer lost or damaged parcels, faster claims.

#08 · IndustrialRoadmap

Ergonomics and Posture Risk

Unsafe lifting, bending and reaching scored from body pose.

Fewer back injuries and lost workdays.

#09 · IndustrialRoadmap

Pharma and Packaging Line Checks

Label presence, cap, fill level and lot code read on every unit.

Recalls prevented before products ship.

#10 · Retail, Food and FitnessRoadmap

Food Safety and Kitchen Hygiene

Hairnets, masks, gloves and handwashing checks in food prep.

Pass audits, protect your brand.

#11 · Agriculture and EnergyRoadmap

Livestock Health Monitoring

Bird and animal counts, activity, crowding and mortality signs.

Earlier disease response, better feed conversion.

#12 · Agriculture and EnergyRoadmap

Aquaculture Fish Counting

Fish counts, size estimation and feeding behaviour underwater.

Feed less waste, know your biomass.

#13 · Construction and MiningRoadmap

Infrastructure Defect Inspection

Road cracks, potholes and structural damage from vehicles or drones.

Repairs planned by evidence, not complaints.

#14 · Agriculture and EnergyRoadmap

Energy and Utility Asset Inspection

Solar panel faults, substation and line defects.

Higher yield, fewer climbs and truck rolls.

#15 · Agriculture and EnergyRoadmap

Precision Agriculture

Crop health, pest signs, tree and fruit counting from drones.

Yield estimates and targeted treatment.

#16 · Retail, Food and FitnessRoadmap

Retail Shelf and Inventory

Empty shelves, misplaced products and planogram checks.

Recover lost sales from stock-outs.

#17 · Transport and LogisticsRoadmap

Port and Container Yard

Container codes, crane moves and yard turnaround measured.

Faster gate turnaround, fewer errors.

#18 · IndustrialPilot

Fire and Smoke Early Warning

Visible smoke and flame flagged in yards and warehouses.

Earlier response alongside certified alarms.

#19 · HealthcareRoadmap

Fall Detection and Elderly Care

Falls and long immobility in care homes and wards.

Staff alerted in seconds, privacy kept.

#20 · IndustrialRoadmap

Equipment Condition Monitoring

Leaks, corrosion, overheating and abnormal visible conditions.

Catch failures before downtime.

#21 · IndustrialRoadmap

Industrial Process Monitoring

Conveyor activity, production counts and material flow.

Live OEE without new sensors.

#22 · IndustrialRoadmap

Visual Measurement and Metrology

Dimensions, clearances and alignment, with calibrated cameras.

Measure without stopping the line.

#23 · Construction and MiningRoadmap

Construction Progress Tracking

Milestones and installed items compared with the plan over time.

Disputes settled with timestamped evidence.

#24 · IndustrialRoadmap

Vision-Guided Robotics

Picking, placing and sorting guided by camera.

Automate repetitive handling.

#25 · Agriculture and EnergyRoadmap

Food Sorting and Grading

Ripeness, size, defects and packaging faults graded.

Consistent grade, less waste.

#26 · Agriculture and EnergyRoadmap

Waste Sorting and Recycling

Materials classified to guide sorting lines.

Purer output, higher resale value.

#27 · IndustrialRoadmap

Intelligent Document Processing

Invoices, delivery orders and forms read and checked.

Data entry cut to review only.

#28 · Transport and LogisticsRoadmap

Insurance Damage Assessment

Vehicle and property damage classified from photos.

Faster, more consistent claims.

#29 · HealthcareRoadmap

Medical Image Support

Radiology and pathology assistance for clinicians.

Second-reader support, not a diagnosis.

#30 · Retail, Food and FitnessRoadmap

Sports and Fitness Analytics

Player tracking, pose and exercise form analysis.

Coaching insight from ordinary video.

Live Running for customers today Pilot Available for a paid site pilot Roadmap On the plan, ask us about timing

Proven on a Live Quarry

Every Truck, Every Load, Counted by Camera.

Our first production deployment watches a granite quarry: a custom-trained YOLO model on Ultralytics Enterprise detects each dump truck, tracks it through the weighbridge and loading bay, and estimates how full the tray is from its volume.

Truck Detection and TrackingSeven custom classes, IDs kept through dust and occlusion.
Payload PercentageLoad fullness estimated per trip, flagged when underloaded.
Trip ReportsTrips per hour, per truck, per shift, in ECharts dashboards.

Excavator loading rock into a dump truck in an open-pit quarry watched by a mast camera

Dump Truck 0.97 · Load 82%

Core Capabilities

Twelve Building Blocks, Thirty Solutions.

Every use case above is a combination of the same twelve vision capabilities. We build each one once and reuse it everywhere, so a new use case is configuration and training, not a new product.

Image Classification

Sorts a whole frame into a category.

Normal / Defective

Object Detection

Finds and labels every object in view.

Person, Car, Helmet

Object Tracking

Keeps one ID per object across frames.

Truck #12 in, out

Segmentation

Marks the exact pixels of each object.

Crack area 4.2 cm²

OCR

Reads plates, container IDs and labels.

Plate, serial, invoice

Counting

Counts items, animals or trips through a line or zone.

Fish per cage

Pose Estimation

Finds body joints and posture.

Worker bending

Action Recognition

Understands movement over time.

Walking, Falling

Anomaly Detection

Learns normal, flags anything else.

Unexpected object

Depth Estimation

Estimates distance and 3D shape.

Forklift 1.8 m away

Visual Measurement

Size, area and volume from calibrated views.

Stockpile volume

Scene Understanding

Describe or search footage in plain words.

"Bunches with long stalks"

How It Works

From Pixels to an Action, in Under a Second.

A fast detector looks at every frame. The heavier vision-language model is only called when something is worth explaining, so you get intelligence without paying GPU prices on every pixel.

Connect

Add line, drone or IP cameras. Draw zones, lines and schedules.

FFmpeg · GStreamer

Detect

Every frame is scanned for the objects your use case needs.

YOLO26 · RF-DETR

Track

Objects keep their ID so we count once and measure dwell time.

ByteTrack · BoT-SORT

Understand

Only flagged clips get a vision-language model to explain them.

VLX-Seek · Qwen3-VL

Decide

Rules turn detections into events: zone, time, count, duration.

IoTServa Rules

Act

Notify, reject a part, stop a conveyor, log a report, all audited.

MQTT · Modbus · Relays

Edge First

Video Stays on Site. Insights Travel.

A small fanless edge box beside your NVR runs the models locally. Only events, snapshots and counts go to the cloud, which saves bandwidth, keeps working when the internet drops and keeps footage under your control.

Runs on Jetson, x86 or ARMTensorRT, OpenVINO or ONNX Runtime, matched to the hardware.
Store and ForwardEvents queue on the box during an outage and sync on reconnect.
One Platform with IoTServaVision events sit beside your sensors, meters and relays.

Fanless edge AI computer mounted on a DIN rail inside a control cabinet

The AI Stack

Best-in-Class Models, Picked per Job.

We are not married to one model. Each layer uses the tool that benchmarks best for your footage, with licences checked before anything ships.

Layer	Tools We Use	Best For
Object Detection	Ultralytics YOLO26, RF-DETR, RTMDet	Fast, accurate boxes on every frame
Open-Vocabulary	Grounding DINO, YOLOE, Florence-2	Find new objects described in words
Segmentation	Meta SAM 2, SAM 3.1, MMSegmentation	Pixel masks, cracks, stockpiles, labelling
Tracking	ByteTrack, BoT-SORT, OC-SORT	Counting, dwell time, identity through occlusion
Vision Language	OM AI VLX-Seek 1.5, Qwen3-VL	Grounded queries and plain-language descriptions
Pose and Action	MMPose, RTMPose, MMAction2	Falls, ergonomics, unsafe lifting
OCR	PaddleOCR, Fast-ALPR, DocTR	Lot codes, labels, container codes, documents
Anomaly	Anomalib (PatchCore, EfficientAD)	Defects with few bad samples
Embeddings	CLIP, SigLIP, DINOv3, Qdrant	Search by image or meaning
Inference	TensorRT, OpenVINO, ONNX Runtime, DeepStream	Many cameras per box at low latency
Data and Training	CVAT, FiftyOne, PyTorch, MLflow	Labelling, error review, retraining on your site

Honest Limits

We Say What Vision Cannot Do.

AI vision is powerful and still imperfect. These are the lines we will not blur in a sales meeting.

Not a Fire Alarm

Smoke and flame detection is an early warning that sits beside certified detectors, never instead of them.

Measurement Needs Calibration

Millimetre claims need calibrated cameras and an uncertainty study. A single photo gives an estimate, not a spec.

Lighting and Lens Matter

Fine defects need the right camera, lens and light. We check optics on site before promising an accuracy number.

Not a Medical Device

Medical imaging features support clinicians and are not a regulated diagnosis. We tell you which items are pilots.

FAQ

Questions We Hear First

Do I need new cameras?

Usually not. Any IP camera with an RTSP or ONVIF stream at 1080p works for most use cases. Fine defects, underwater and long-range views need the right lens, lighting and mounting, which we check during the site assessment.

Does video leave my site?

It does not have to. Inference can run on an edge box on site, sending only events, snapshots and counts to the cloud. Cloud processing is optional.

How accurate is it?

It depends on the camera angle, lighting and how close your site is to the training data. We measure precision and recall on your own footage during the pilot and share the numbers before you commit.

Which AI models do you use?

We pick per use case: YOLO-family and RF-DETR detectors, ByteTrack or BoT-SORT tracking, SAM segmentation, PaddleOCR, Anomalib for defects, MMPose for posture and vision-language models for natural-language search, fine-tuned on your footage where accuracy matters.

How does it connect to my other systems?

AiVisionBot runs on the IoTServa platform, so a vision event can trigger anything IoTServa controls: email, WhatsApp or app alerts, relays, reject gates, conveyors, MQTT, Modbus and the Open API.

Can it replace a certified fire alarm?

No. Visual smoke and fire detection is an early-warning layer that adds to, never replaces, certified detection and alarm systems.

Show Us One Camera. We Will Show You What It Can See.

Send a short clip from your site and pick a use case. We run it through AiVisionBot and walk you through the detections, the accuracy and the cost before you buy anything.

Request a Demo Browse Use Cases