Computer Vision for Manufacturing: What to Ask Vendors  at GITEX Global 2026 

If you run a plant, a warehouse or a dispatch yard, at least a dozen exhibitors at GITEX Global 2026 will  offer you computer vision. Most will show a polished demo. Very few will explain what happens on the  night shift when the lighting changes, a new product variant arrives and the supervisor’s phone fills with  false alerts. 

Computer vision for manufacturing means using cameras and software to see, count, inspect and verify  what happens on the factory floor, then turning that into records, alerts and decisions your team can act  on. This guide is written for the buyer. It covers what to prepare before GITEX, twelve questions that  separate real deployments from demos, the red flags to watch for, and how to book a meeting with  Easycomm in Dubai from 7 to 11 December 2026. 

Key takeaways 

• Define one measurable problem before you walk the halls: miscounts, wrong dispatches, PPE gaps  or defects. 

• Ask for false-positive and false-negative rates, not a single accuracy figure. 

• Insist on testing with your own footage, or on a pilot you scope yourself. 

• Check how detections reach your ERP, WMS or supervisors, and who reviews them. • Ask where video is stored and how worker privacy is handled under UAE law. 

• Use the same twelve questions in every meeting so vendors can be compared fairly. 

What computer vision in manufacturing actually does 

Computer vision turns video from cameras into structured information. On a production floor or in a  warehouse, that usually comes down to five jobs: 

• Counting: bales, cartons, pallets, sheets and units at inward, pressing, packing and loading. 

• Verification: matching what is physically loaded against an order or invoice before a truck leaves,  often called dispatch verification. 

• Safety and compliance: PPE detection, restricted-zone entry, forklift and pedestrian proximity, and  fire and smoke detection. 

• Inspection: damaged cartons, bent or torn sheets, loose straps and visible defects.

• Tracking: vehicles at the gate, pallet movement between areas and inventory levels. 

The value is not the detection itself. It is what happens next: an alert to the right person, an evidence  image, and a record reconciled with your business data. Easycomm describes the flow as see, understand,  act. Cameras identify what is happening, software compares it with what should be happening, and a  person reviews anything that matters. Human review is a design choice, not a weakness: the system  supports supervisors and does not replace their judgement. 

You can see how this applies to plants in our AI video analytics for manufacturing page, and to docks and  yards in our warehouse and logistics video analytics page.

Why GITEX Global 2026 matters for manufacturing buyers 

GITEX Global 2026 opens with a Summit on 7 December, followed by a four-day expo from 8 to 11  December at the Dubai Exhibition Centre, Expo City Dubai. It is the first edition at this venue. Organisers  expect more than 200,000 attendees and around 6,800 exhibiting companies from over 180 countries. 

For manufacturing and logistics buyers, three parts of the 2026 programme stand out. The Advanced  Manufacturing zone covers robotics, automation, digital twins and sustainable supply chains. The Physical  AI and Robotics Park focuses on intelligent machines, including drones, autonomous systems and industrial  automation. AI Everything continues to centre on artificial intelligence, with agentic AI as a 2026 focus. 

That scale is useful and risky at the same time. You can compare many vendors in a few days, but without a  plan you will collect brochures instead of answers. Our GITEX Global 2026 guide covers dates and logistics,  and our meeting strategy for finding the right tech partner explains how to structure your week. 

Before you go: three things to prepare 

1. Name the problem in one sentence and attach a number. How many miscounts, wrong loads or PPE  gaps do you see per month today? A vendor can only prove value against a baseline you can measure. 

2. Map your camera and system landscape. Note how many cameras you have, which brands, which  recorder or video management system is in use, which ERP, WMS or MES holds your records, and how  much network capacity each site has. 

3. Decide what success looks like. Choose the one metric a pilot must move, such as fewer dispatch  mismatches or faster dock checks, and write down how many false alerts your team can tolerate. Buyers  who set the metric themselves get pilots that reflect real operating conditions. 

12 questions to ask every computer vision vendor at GITEX 

Ask these in writing where you can, and ask for evidence behind each answer, not just a yes. Each question  below includes why it matters and what a good answer looks like. 

Proof and accuracy 

Question 1: Which exact problem have you solved, and how did you measure it? 

Computer vision covers dozens of different jobs. A vendor that counts pallets at a dock has not necessarily  inspected surface defects. A good answer names the use case, the metric used and a deployment similar to  yours. 

Question 2: What are your false-positive and false-negative rates in conditions like mine? 

A single accuracy number hides the trade-off. As an illustration, a line packing 10,000 cartons per shift with  a 2% false-reject rate flags 200 good cartons every shift, and operators soon stop trusting the alerts. For  safety use cases, missed events matter even more. Ask for both numbers at a stated confidence threshold,  and ask what changes when the threshold moves. 

Published results show how context-specific these numbers are. A 2022 research paper on counting bales  and product boxes on an industrial line reported 99.91% accuracy in its own setup, and a 2022 paper on  PPE monitoring reported 0.93 precision and 0.96 recall in its evaluation. Neither guarantees the same result  in your plant. A good answer gives figures from a comparable site, along with how many cases were  measured and over what period.

Question 3: Can you test on my own footage, or on my site, before I commit? 

Demos are usually run in controlled conditions. A good answer is a willingness to process a sample of your  recordings, including the difficult shifts, and to report results against a count you trust. 

Question 4: How does the system cope with change: lighting, occlusion and new product variants? 

Plants change constantly. Ask exactly what happens when a new SKU, a new pack design or a new camera  angle arrives, who retrains the model and how long it takes. A good answer describes a defined retraining  process with a turnaround time and clear responsibility. 

Integration and deployment 

Question 5: Will it work with the cameras and recorders I already have? 

Replacing cameras can cost more than the software. Ask about ONVIF and RTSP support, compatibility with  your recorder or video management system, analog cameras and image requirements such as resolution,  angle and lighting. A good answer includes a site survey that confirms which cameras qualify and which  need repositioning. 

Question 6: How do detections reach my ERP, WMS, MES or dashboards? 

A count that sits in a separate dashboard becomes a second system of record. Ask about APIs, webhooks  and data formats, and whether detections can be reconciled with orders, invoices and lots. If you are  connecting vision data to systems you already own, our guide on the cost of AI integration for existing  business software explains what to budget. A good answer includes documented APIs and a reference  integration with a system like yours. 

Question 7: Where does processing happen: edge, on-premises, cloud or hybrid? 

This choice affects latency, bandwidth, cost and data privacy. Plants with weak connectivity often need  local processing. A good answer explains the options and shows how each one changes cost and response  time. 

Operations and people 

Question 8: What happens after a detection, and who reviews it? 

Alerts without a workflow create noise. Ask who receives the alert, how an evidence image or short clip is  attached, how false alerts are marked and fed back to improve the system, and how events escalate. A  good answer includes a human review step and an audit trail, particularly for anything involving workers. 

Question 9: How will the pilot be scoped, measured and ended? 

Pilots fail when the vendor controls scope alone or when success is vague. Agree one metric, a duration,  the cameras involved, acceptance criteria and what happens if the target is missed. A good answer is a  written pilot plan that you co-author, with a go or no-go decision date. 

Question 10: Who supports the system after go-live, and how quickly do they respond? 

Cameras move, firmware updates break streams and models drift. Ask about monitoring, camera-health  alerts, response times and the update process. Ask for two references in your industry with similar site size,  and ask each one what went wrong and how it was fixed. A good answer is a named support process and  honest references.

Privacy, compliance and cost 

Question 11: How is video data stored, secured and governed, including worker privacy under UAE law? 

In onshore UAE, Federal Decree-Law No. 45 of 2021 on personal data protection applies, and it has been in  force since 2 January 2022. Footage that identifies people counts as personal data. Legal commentators  advise that workplace monitoring should be lawful, proportionate and transparent, with staff informed.  Free zones such as DIFC and ADGM have their own data protection regimes. Ask about retention periods,  access controls, masking, hosting location and what the vendor expects you to put in place, such as notices  and policies. This is not legal advice, so confirm your obligations with qualified counsel. A good answer is  written data-handling documentation and a clear split of responsibilities. If you are new to the market, our  overview of a software development company in Dubai explains what UAE businesses should consider. 

Question 12: What does it really cost, and what happens if I want to leave or scale? 

Pricing may include per-camera fees, per-use-case fees, hardware, integration, retraining and support. Ask  for the first-year cost and the cost at full scale. Ask who owns the models, configuration and recorded data,  and how you can export them. A good answer is an itemised quote with clear exit terms. 

Red flags to watch for at the booth 

• A single “99% accuracy” number with no split between false positives and false negatives. • Demo videos only, with no willingness to run your footage. 

• A pilot scoped entirely by the vendor, or a success metric that stays vague. 

• “Works with any camera” without a site survey. 

• No clear answer on where video is stored or who can access it. 

• Worker-related detections presented as automatic violations, with no human review step. • No references in your industry, or no honest account of a deployment that did not go smoothly. 

The same discipline applies when you choose any technology partner. Our lists of the top 10 questions to  ask before hiring a software development company and the points to evaluate when choosing an AI  development company are useful companions. 

Where computer vision pays back first on the factory floor 

Start with a use case that has a clear, countable outcome. Based on Easycomm’s work across  manufacturing, warehousing, agro-processing and packaging operations, these are practical starting points: 

Counting and reconciliation. Cameras at each count point compare what they see with what the record  expects, so a short count is flagged before dispatch instead of after a customer complaint. In one  engagement with a cotton and hygiene products manufacturer in central India, this covers bales at raw  material inward, boxes at pressing, SKU counts at packing and the loading count at the dock. 

Dispatch verification. In a palletised-carton warehouse operation, each pallet receives a digital identity at  receipt. Cameras follow it through storage, forklift movement and loading, then validate the load against  the dispatch order. Every discrepancy carries an evidence image and a short video clip. 

Safety visibility. PPE and restricted-area monitoring, fire and smoke detection and vehicle identification at  the gate. Each detection becomes an event for a person to confirm, never an automatic violation.

Packaging and bundle inspection. For a cardboard sheet manufacturer, each finished bundle is registered,  counted and inspected for bent, torn or folded sheets and loose straps, then tracked until it is verified at  dispatch, leaving an image-backed history of every bundle. 

Results depend on your site, your cameras and your data. Ask any vendor, including us, for measured  results from an environment that resembles yours. 

Our Computer Vision Work: Four Real Deployments

We have already worked on real computer vision problems across manufacturing, warehousing, agro-processing and packaging. Each engagement started from an operational problem, not from a technology. In every case, cameras act as a physical verification layer: they confirm what actually happened and flag any gap against the expected record, and every detection that matters is routed to a person for review. The system supports supervisors; it does not replace their judgement.

1. Mohini Health & Hygiene Ltd, Pithampur (Madhya Pradesh)

Challenge: bale, box, loading and SKU counts depended on manual checks, so a short count at inward, pressing or loading could go unnoticed until a customer or stock mismatch surfaced.

• Counting bales at raw material inward, boxes at pressing, loading at the dock and SKUs at packing
• Vehicle identification at the main gate against expected trucks
• PPE, restricted-area entry, fire, smoke and smoking detection, each confirmed by a person before it becomes an event

Designed to: reconcile every count with the business record and stop or flag dispatch when the box count, product or order does not match.

2. Zimgold, Zimbabwe

Challenge: counting, product identification and loading checks were manual, so wrong-truck, short or damaged dispatches were hard to prevent and harder to prove afterwards.

• A digital identity for each pallet at receipt, tracked through storage, forklift movement and loading
• Product, label, carton and pallet identification and counting
• Pallet-to-truck matching and detection of torn, wet or open cartons and missing labels

Designed to: check expected versus detected SKU, quantity and truck before dispatch, with an evidence image and short video clip for every discrepancy.

3. Avi Agri Business Pvt. Ltd. (Aviagri), near Ujjain (Madhya Pradesh)

Challenge: limited visibility of inventory and of employee activity on site.

Approach: a camera-based solution covering inventory monitoring (levels and movement) and employee presence and activity. Designed to: reduce dependence on manual stock checks, surface discrepancies early and give supervisors a factual view of on-site activity.

4. Pack Universe, finished-goods warehouse

Challenge: sheet counts per bundle and bundle quality were checked by hand, so short counts, damaged sheets and unstable bundles could reach the customer.

• Registering each new bundle and its storage location
• Counting sheets against the expected quantity
• Detecting bent, torn, folded or crushed sheets, loose straps, gaps and uneven stacking

Designed to: cut manual counting and inspection, speed up dispatch checks and keep an image-backed history of every bundle.

Want to know whether a similar setup fits your plant or warehouse? Talk to our team about scope and pricing, or see our AI video analytics for manufacturing.

What Easycomm brings to the conversation 

Easycomm Innovations builds computer vision and AI systems together with the software around them. On  production floors, our platform runs on existing ONVIF-compatible cameras and recorders, with PPE  detection, restricted-zone and line-crossing alerts, person and vehicle detection, loitering detection and  real-time alerting. For warehouses, it adds forklift and pedestrian proximity alerts, intrusion detection,  licence plate recognition and trailer dwell-time tracking, with integration into warehouse and yard systems  through APIs. 

Behind the platform, our AI software development team builds custom computer vision pipelines for  object recognition and counting, defect detection and real-time video analysis. We also build ERP and CRM systems so detections land in the tools your teams already use, and AI agents that can act on events. Our  team expects to be asked all twelve questions above, and we would rather you ask them. 

Meet Easycomm at GITEX Global 2026 

Easycomm will be at GITEX Global 2026 in Dubai from 7 to 11 December. Bring your problem statement,  your camera list and the twelve questions above. We will walk through whether computer vision fits your  use case and what a scoped pilot could look like. You can read more about our plans in Easycomm at GITEX  Global 2026 and see the full list in our 11 technology solutions to discuss at GITEX. 

• Book a founder meeting: Book a meeting with Easycomm at GITEX. Meeting slots run between  10:00 AM and 5:00 PM GST, and we confirm your slot by email. 

• Prefer to pick a time yourself? Schedule directly on Calendly. 

• Not attending GITEX? Use our contact page and we will arrange a call. 

Beyond computer vision, we can also discuss AI CCTV, AI hybrid solutions, robotics built on the ROS stack,  web and app development, AI development and AI agents, real estate software with a built-in CRM, custom  software, an LMS with AR and VR features, ERP and CRM. 

Conclusion 

Computer vision for manufacturing works when it starts from a measurable operational problem, is tested  on your own conditions and is integrated with the systems your team already uses. GITEX Global 2026 gives  you a rare chance to compare vendors side by side. Walk in with a baseline, ask the twelve questions, and  judge every vendor on evidence rather than demos. 

Ready to talk? Book a founder meeting with Easycomm at GITEX Global 2026 in Dubai, 7 to 11 December.

Sources 

• GITEX Global 2026 official page: https://www.gitex.com/gitex-global-2026 

• GITEX 2026 zones (CX Today): https://www.cxtoday.com/events/gitex-global-2026/

• GITEX 2026 programme (The Recursive): https://www.therecursive.com/gitex-global-2026-moves to-expo-city/ 

• GITEX 2026 attendance expectations (Egypt Innovate): https://egyptinnovate.com/en/ehttps://egyptinnovate.com/en/events/gitex-global-2026vents/gitex global-2026 

• UAE data protection and employee monitoring (Freshfields):  

https://riskandcompliance.freshfields.com/post/102hk68/uaes-new-data-protection-law-provides more-transparency-and-enhanced-rightsmean 

• CCTV rules in Dubai (Meydan Free Zone): https://www.meydanfz.ae/blog/cctv-rules-dubai

• Computer vision vendor due diligence (Protex): https://www.protex.ai/post/computer-vision-vendor-due-diligence—12-questions-it-should-ask-before-a-pilot

• Measuring computer vision pilot success (Edge AI and Vision Alliance): https://www.edge-ai vision.com/2022/06/computer-vision-in-manufacturing-the-definitive-guide-to-measuring-the success-of-a-pilot/ 

• Bale and product box counting paper (ICCSA 2022): https://bv.fapesp.br/en/publicacao/222536

• PPE safety monitoring paper (IEEE, 2022): https://ieeexplore.ieee.org/document/10065769

• Easycomm manufacturing video analytics page: https://easycomm.io/video-analytics-software/manufacturing/

• Easycomm logistics video analytics page: https://easycomm.io/video-analytics-software/logistics/

• Easycomm AI software development page: https://easycomm.io/services/ai-software development/ 

• Easycomm GITEX booking page: https://easycomm.io/gitex-global-2026/