FAQs
Frequently Asked Questions
Everything you need to know about getting started with XenReality's Vision AI solutions.
General
Do I need to install specialized cameras?
In most cases, no. XenReality's solutions can work with existing camera infrastructure where suitable. For example, XenTrack can use existing CCTV feeds. For best accuracy, cameras should have a minimum resolution of 4MP and provide suitable coverage, positioning, lighting, and image quality.
How accurate are XenReality's solutions?
Accuracy depends on factors such as camera placement, lighting, and how busy the environment is. XenReality's solutions typically achieve 90–95% accuracy, and performance can be validated during a pilot using the customer's own site conditions.
Does video leave the premises?
No. All processing runs on a device installed on-site, with only the resulting metrics displayed on a unified dashboard for the customer. If the customer chooses a cloud-based solution, data can instead be stored and processed in the cloud.
Can XenReality's metrics integrate with existing dashboards?
Yes. XenReality's solutions are designed to integrate with existing business systems and workflows through APIs and other integration options, depending on the specific product and use case.
What is required to get started?
Typically, the first step is to discuss the use case and site requirements with the XenReality team. A pilot can then be deployed to validate performance on the customer's own environment before a wider deployment. A proof of concept can be done in under a week.
Vision AI Basics
What is Vision AI?
Vision AI, also called computer vision, is software that uses machine learning to understand images and video. It can detect objects, count people, read text and spot defects, turning camera footage into structured data a business can act on.
How is Vision AI different from ordinary video surveillance?
Traditional CCTV records footage for people to review later. Vision AI analyses the footage automatically and continuously, producing counts, alerts and reports, so teams only need to look at what matters.
What is edge AI and why does it matter?
Edge AI runs the models on a device at your site, close to the camera, instead of sending video to a remote server. It reduces delay, keeps footage on-site and lowers bandwidth needs, which is why it suits real-time uses such as production lines.
What is the difference between on-premises, private cloud and public cloud deployment?
On-premises means the system runs on hardware at your own site. A private cloud is a dedicated environment that your organisation controls. A public cloud uses shared infrastructure from a provider. XenReality supports on-premises and private cloud deployments, so data can stay within your own environment.
How is a Vision AI model trained?
Models learn from labelled examples: images or video where the things of interest, such as defect types or objects, are marked. The model learns the patterns and is then tested on separate examples it has not seen. The more representative the examples are of your real site, the better it performs.
What affects the accuracy of computer vision?
Camera placement and resolution, lighting, crowding or objects blocking the view, motion blur, and how much the things being detected vary all play a part. The quality and variety of the training examples matter too. Testing on your own footage is the most reliable way to know what to expect.
What are false positives and false negatives?
A false positive is when the system flags something that is not there, such as a defect on a good product. A false negative is when it misses something that is there. The two trade off against each other, so the right balance depends on your use case and is tuned and validated with your own data.
What is a pilot or proof of concept and why start with one?
It is a small trial on your own cameras, images or documents to measure real performance before you scale up. It shows what accuracy and outputs you can expect in your conditions, and lets you decide with evidence rather than assumptions.
Is computer vision compatible with data privacy laws?
It depends on what is being analysed and how. Counting people or inspecting products does not require identifying individuals, while face recognition is subject to stricter rules and consent requirements. Laws such as the EU's GDPR and India's Digital Personal Data Protection Act, 2023 may apply, so involve your legal team early. XenReality's privacy policy follows GDPR principles.
What is the difference between OCR and intelligent document processing?
OCR converts the text in an image into machine-readable text. Intelligent document processing goes further: it identifies the type of document, finds specific fields such as an invoice number or total, and checks them, so the result can flow straight into your systems.
How long does a deployment take?
It depends on the use case and site, but XenReality aims to take customers from contract to live production in under four weeks. A pilot is usually the first step so performance is confirmed on your own site before a full rollout.
Which industries use Vision AI?
Common examples include retail and shopping centres, airports and transport hubs, stadiums and live venues, hotels and restaurants, manufacturing and automotive, electronics, food and beverage, pharmaceutical packaging, logistics, energy, finance and insurance, education, car parks and smart cities.
Deployment & Integration
How do I connect XenReality to my existing cameras?
Most IP cameras and CCTV systems can be used by reading their video streams over the network, typically through standard protocols such as RTSP. Share the camera make, resolution and mounting positions with the team and they will confirm what is needed before any work starts.
Do I need new hardware to run XenReality?
Usually a small processing device is installed on-site to run the models next to your cameras, while the cameras themselves are reused where they are suitable. The exact hardware depends on how many cameras you have and which products you use.
What happens if my internet connection goes down?
When processing runs on-site, the analysis keeps running locally even if the internet connection drops, which is one of the advantages of running AI at the edge. How and when results reach dashboards depends on your setup, so confirm this with the team.
Can XenReality integrate with ERP, MES, CRM or BI tools?
Outputs are structured, so they can be sent to other systems through integrations and APIs, or shown in dashboards. Share which systems you use and the team can advise on the simplest way to connect them.
Can one deployment cover multiple sites?
Yes. Many customers start with one site or one line, confirm the results, and then repeat the same setup elsewhere. Results from different sites can be compared once they are brought into a common view.
How many cameras do I need?
It depends on the area to be covered and the question you want answered. A single entrance may need one camera, while a large store floor or a full production line needs several. A site review gives a firm number.
Who maintains the system after it goes live?
Models can be monitored and improved over time as more data is collected. Support and update arrangements are agreed with you during the initial discussion, so you know who is responsible for what.
Can the system be customised for my specific use case?
Yes. XenReality builds custom models trained on your own products, environment and requirements, so accuracy is not limited to generic examples.
Security & Data Privacy
Where is my video and image data stored?
With on-premises or private cloud deployment, data stays within your own environment. Only the results, such as counts, alerts and reports, need to be shown on dashboards. If you choose a cloud-based setup, data is stored and processed in the cloud instead.
Is the data encrypted?
XenReality describes its security as enterprise-grade, with data sovereignty and compliance built in. Ask the team for the specific controls, including encryption, that apply to your chosen deployment.
Who can access the dashboards?
Access is limited to the people you nominate. Ask the team about the access controls available for your deployment, so different teams only see what they need.
Does XenReality use my data to train its models?
XenReality's policy is that customer data is not used to train its AI models without explicit written consent.
How long is footage kept?
Retention is agreed with you and can be set to match your own policy and legal requirements. Many analytics uses only need the results, so raw video does not have to be kept.
Can analytics be done without identifying individuals?
Yes. Counting people, measuring dwell time, mapping movement and inspecting products do not require knowing who anyone is. Identification, such as face recognition, is a separate use case with stricter requirements.
Retail & Space Analytics
What is footfall and why does it matter?
Footfall is the number of people who enter a store or area. It is the starting point for understanding performance because sales, staffing and marketing can only be judged fairly against how many people actually visited.
What is dwell time?
Dwell time is how long visitors stay in an area. A long dwell time in a display zone suggests interest, while very short visits can show that a zone is being passed by.
How can footfall data help me schedule staff?
Counts by hour and day show when the store is busiest. You can then match staffing to real demand instead of fixed rosters, so customers are served at peak times without overstaffing quiet periods.
How can I tell if a promotion or display is working?
Compare footfall and dwell time in the zone before, during and after the change. If more people stop and stay longer, the display is attracting attention, and you can compare that with sales for the same period.
What is queue analytics?
It measures how many people are waiting and for how long. It can be used to flag when a queue passes a limit you define, so another till can be opened before customers lose patience.
Can people counting handle crowded entrances?
Accuracy depends on camera position and how busy the entrance is. Cameras mounted high and looking down reduce people blocking each other, and a pilot on your own entrance shows what to expect in your conditions.
Manufacturing & Quality Control
What is automated visual inspection?
It uses cameras and AI to check products for defects automatically as they move along a line, flagging or rejecting items that fail, and recording what was found so quality can be tracked.
What kinds of surface defects can computer vision find?
Common examples are scratches, dents, cracks, discolouration, stains, missing parts and misalignment. What is practical depends on image quality, lighting and the examples the model is trained on.
Does inspection slow my production line?
It is designed to keep pace with the line. XenInspect runs optimised models at line speed with sub-second inference, so inspection happens without slowing production.
How is lighting handled for inspection cameras?
Consistent lighting makes inspection more reliable, so it is often planned together with the camera. Models can handle some variation, but stable lighting gives the most dependable results.
Can inspection results be traced to a batch or shift?
Yes. Results can be summarised per batch and per shift, with defect types and image evidence, which supports traceability and root-cause analysis upstream.
What is the typical return from automating inspection?
Savings usually come from fewer escaped defects, less rework and scrap, and less manual checking time. The size of the return depends on your line, defect rate and current process, so it is best estimated after a pilot.
Document Automation
What documents can be digitised with OCR?
Invoices, purchase orders, receipts, forms, labels, ID documents and printed reports are common examples, as long as the images are clear enough to read.
How accurate is OCR on scanned documents?
It improves with higher resolution, even lighting and straight, clean scans. Handwriting, stamps and low-quality photos are harder, so results should be checked on your own sample documents.
Can OCR handle tables and line items?
Yes. Beyond reading text, extraction can recognise table structure so that items, quantities, rates and totals are captured in the right columns.
How do I reduce errors in manual data entry?
Automating capture removes most retyping, and confidence scores can send only uncertain fields to a person for review. That keeps accuracy high while saving time.
Can extracted data flow into my accounting or ERP system?
Extracted fields are structured, so they can be passed to other systems through integrations or exported in standard formats. Share your system and the team can advise on the simplest route.
Does OCR support multiple languages?
Language support depends on the setup. If your documents use more than one language or script, share samples so suitability can be confirmed before you start.
Recognition & Access Control
What is automatic number plate recognition used for?
Common uses include car park management, gated entry, toll and traffic monitoring, and logging vehicle movements on a site.
How does automatic number plate recognition work?
A camera captures the vehicle, the system finds the plate, reads the characters and compares them with a list, or simply logs them, depending on how it is set up.
Can recognition run on existing cameras?
In many cases, yes, if the camera angle, resolution and lighting suit the task. A site review confirms whether the cameras already in place are good enough.
How accurate is number plate recognition at night?
Night performance depends on lighting and camera quality, particularly glare from headlights. Infrared-capable cameras and good positioning help, and a pilot shows what to expect on your site.
What is a watchlist and how is it used?
A watchlist is a list of people or vehicles that trigger an alert when detected. It is used for example to flag unauthorised vehicles or alert security staff, and it should always have a clear legal basis and policy.
How can recognition support access control?
It can open a gate or log an entry when a recognised vehicle or person arrives, replacing manual checks or cards. Time-based and zone-based rules decide who may enter and when.
XenInspect - AI Visual Inspection
Explore XenInspectWhat types of defects can XenInspect detect?
XenInspect is designed to flag faults and defects on production lines in real time. The system can be customized for the customer's products and defect types, with the specific inspection requirements assessed for each use case.
How does automated visual inspection compare with manual QC?
XenInspect provides production-speed defect detection and automated QC reporting, helping teams identify faults consistently during production rather than relying only on manual inspection. Compared with manual QC processes, XenInspect can deliver up to 30% time savings and up to 50% cost savings, while providing 100% visual inspection traceability.
Can XenInspect be used on an existing production line?
Yes. XenInspect is designed to integrate directly with production lines. The setup typically requires suitable camera placement and coverage of the inspection area, appropriate lighting for clear and consistent images, and integration with the production-line workflow. XenInspect supports multi-camera setups for 360° coverage and is designed to operate at full production speed.
How is inspection accuracy validated?
Inspection accuracy should be validated using representative samples from the customer's actual production line, including both good products and known defect types. The system's results should then be compared against the customer's existing/manual QC results to measure detection accuracy, false positives and false negatives under real production conditions. A pilot can be used to establish and validate performance before wider deployment.
What is the difference between rule-based machine vision and AI-based visual inspection?
Rule-based machine vision checks fixed measurements and thresholds, which works well for simple, consistent parts. AI-based inspection learns from examples, so it can handle natural variation in finish, lighting and defect appearance that fixed rules struggle with.
How do you add a new defect type to an inspection model?
New defect types can be added by showing the model examples of the new defect. XenInspect supports few-shot learning, so a small number of good examples can often be enough to start, then accuracy improves as more data is collected.
XenRead - Document & Image OCR
Explore XenReadWhat types of documents can XenRead process?
XenRead can process images and documents including forms, invoices, labels and other business documents. It supports both handwritten and printed data processing.
Can XenRead extract data from invoices and forms?
Yes. XenRead extracts, validates and routes information from invoices, forms, labels and images, with field-level extraction supported through its OCR capabilities.
Can extracted data be integrated with existing systems?
Yes. XenRead provides a REST API and webhook delivery, allowing extracted information to be connected to existing workflows and systems.
How can XenRead reduce manual data entry?
XenRead converts information from images and documents into structured data, helping reduce the need for teams to manually read, transcribe and route information.
What is the difference between OCR and XenRead?
OCR on its own turns an image of text into characters. XenRead builds on that to extract the information that matters from documents and images and return it in a structured form you can use in your workflow.
Can XenRead read poor-quality or handwritten documents?
Results depend on how legible the source is, including resolution, lighting, skew and handwriting style. Share representative samples with the team and they can assess suitability for your documents before you commit.
XenTrack - Footfall & Space Analytics
Explore XenTrackWhat does XenTrack measure?
XenTrack provides real-time footfall counting, zone-level dwell-time and engagement metrics, people movement tracking, and AI-generated heatmaps. It also provides dashboards and reports for actionable insights.
What is the difference between footfall and passerby count?
Passerby count refers to people who walk past a location, while footfall refers to people who enter the space. Comparing the two helps businesses understand how effectively a location converts passing traffic into visitors.
Can XenTrack measure customer dwell time?
Yes. XenTrack provides zone-level dwell-time analytics, allowing businesses to understand how long people spend in different areas of a space.
Can XenTrack help with queue and occupancy analytics?
XenTrack provides people movement, footfall and zone-level engagement metrics that can help teams understand busy periods and how people move through a space. Specific queue or occupancy requirements can be evaluated for the customer's use case.
How does camera-based footfall counting work?
A camera watches a defined entrance or area and the system detects each person and counts them as they cross a line or enter a zone. Because it analyses video, it can also follow movement inside the space to report dwell time and popular areas.
What is a retail heatmap?
A heatmap shows where people spend time in a space by colouring areas according to how often they are visited or how long people stay. It helps teams see which displays, aisles or zones attract attention and which are overlooked.
How do I calculate store conversion rate from footfall?
Conversion rate is the number of transactions divided by the number of visitors in the same period. Accurate footfall counts make that figure more reliable than estimates or manual tallies.
XenScan - Recognition & Access
Explore XenScanWhat is XenScan used for?
XenScan provides facial recognition, object tracking and number plate recognition for security, operational intelligence and access-control use cases.
What is ANPR and how can XenScan be used for it?
ANPR, or Automatic Number Plate Recognition, identifies vehicle number plates. XenScan can support vehicle tracking across entry points, car parks and roadways.
Can XenScan support access control?
Yes. XenScan is designed for access-control use cases and provides facial recognition, number plate recognition and watchlist alert capabilities.
Can XenScan work across multiple camera feeds?
Yes. XenScan supports multi-feed facial recognition and vehicle tracking across multiple locations such as entry points, car parks and roadways.
What is the difference between face detection and face recognition?
Face detection only finds that a face is present in an image. Face recognition goes further and matches the face to a known identity, which brings stricter privacy and consent requirements.
What is the difference between ANPR, ALPR and LPR?
They all refer to reading vehicle number plates automatically. ANPR (automatic number plate recognition) is the common term in India and the UK, while ALPR and LPR are used in North America.
Do I need consent to use facial recognition?
Rules vary by country and by use. In many places, matching faces to identities requires a clear legal basis and often consent, plus a notice to people who are affected. Take legal advice for your location and use case before deploying.
Still have questions?
Tell us about your use case and we'll get back to you within 24 - 48 hours.