Quality inspection
Detecting defects, verifying assembly and measuring dimensions.
Technology / Perception
Computer Vision
Computer vision gives software the ability to interpret images and video. It is used to inspect products, read documents, monitor sites and measure objects. Its success depends heavily on real-world conditions, which is why careful data collection matters as much as model choice.
What is Computer Vision?
Short answer
Computer vision is a field of artificial intelligence that trains computers to interpret and understand visual information from images and video, performing tasks such as image classification, object detection, segmentation, tracking, optical character recognition, pose estimation and visual inspection.
Modern computer vision relies mainly on deep learning. Convolutional neural networks and vision transformers learn visual features from labelled images, and pre-trained models can be adapted to specific tasks with relatively modest data. Vision language models add the ability to describe and answer questions about images.
Deployments range from cloud services processing uploaded images to edge devices running real-time detection beside a production line. Practical results depend on lighting, camera placement, image quality, variation in what is being observed and the rarity of the events to detect.
Related concepts
- Object detection
- Segmentation
- OCR
- Edge AI
- Multimodal AI
- IoT cameras
- Deep learning
How it works
How does Computer Vision work?
-
Capture
Cameras or scanners capture images or video under specific conditions.
-
Pre-process
Images are resized, normalised and sometimes enhanced.
-
Model inference
A trained model classifies, detects, segments or reads content.
-
Post-process
Detections are filtered, counted, measured or tracked across frames.
-
Act
Results trigger alerts, update systems or are reviewed by people.
Applications
Where is Computer Vision useful?
Common applications across industries. Suitability always depends on the specific data, process and constraints.
Document reading
Extracting text and layout from scanned documents and forms.
Retail analytics
Shelf availability, planogram checks and anonymous footfall counts.
Agriculture
Crop health, pest detection and produce grading.
Infrastructure inspection
Detecting cracks and corrosion from drones or fixed cameras.
Safety monitoring
PPE and restricted zone detection as support for safety teams.
Limitations
Limitations and when not to use it
Every technology has boundaries. Knowing them early avoids expensive disappointment.
- Accuracy drops with poor lighting, occlusion, motion blur or unusual angles.
- Rare defects or events are difficult to learn with few examples.
- Models trained at one site may not transfer to another.
- Video analytics can raise privacy and surveillance concerns.
- High-resolution, real-time processing requires suitable hardware.
Common misconceptions
Any camera will do.
Resolution, placement, lighting and frame rate often decide whether a use case works.
Vision models work everywhere once trained.
Performance must be validated in each environment and monitored over time.
Computer vision means facial recognition.
Most business vision tasks involve objects, products, documents and scenes, not identifying people.
Comparison
Common computer vision tasks
Choosing the right task formulation affects data labelling, accuracy and cost.
| Task | Output | Example |
|---|---|---|
| Classification | A label for the whole image | Is this part acceptable or defective? |
| Object detection | Boxes around objects with labels | Where are the products on this shelf? |
| Segmentation | Pixel-level regions | Exactly which area of the surface is corroded? |
| OCR and layout | Text and document structure | What does this invoice say? |
| Tracking | Object identities across frames | How did items move through the warehouse? |
Decision guide
How to evaluate Computer Vision for your organisation
Questions worth answering before investing. They form the basis of our feasibility and proof-of-concept work.
| Question to ask | Why it matters |
|---|---|
| Can the target be seen clearly by a camera? | If people struggle to see it in images, models will too. |
| How variable are conditions? | Lighting, weather and product variety increase data needs. |
| How rare are the events of interest? | Rare events need targeted collection or anomaly approaches. |
| Does it need real-time results? | Determines edge versus cloud architecture. |
| Are there privacy implications? | Affects design, legal review and data retention. |
AARHIT services
Services that apply Computer Vision
How we put this technology to work for organisations.
Hybrid Intelligence
Computer Vision Solutions
Detect, classify, measure and inspect using images and video, from cloud pipelines to on-device inference.
Explore service Advance TechnologyInternet of Things and Edge AI
Connect devices and sensors, and run inference close to where data is produced when latency or connectivity demands it.
Explore service AutomationIndustrial Automation
Connect plant systems, sensors and control data with analytics and AI for monitoring, quality and maintenance.
Explore service Hybrid IntelligenceMultimodal AI Solutions
Build systems that reason across text, images, audio, video, documents and sensor data together.
Explore serviceRelated technologies
Industries
FAQ
Computer Vision FAQs
What is computer vision used for?
Quality inspection, document reading, retail analytics, agricultural monitoring, infrastructure inspection, safety monitoring, medical imaging support and many other tasks that involve interpreting images or video.
How accurate is computer vision?
Accuracy varies widely by task and conditions. Controlled inspection with good lighting can be highly accurate, while variable outdoor scenes are harder. Performance should be measured on data from the real environment.
What is the difference between object detection and image classification?
Classification assigns a label to a whole image. Object detection finds and labels individual objects within an image, with their locations.
Can computer vision run on edge devices?
Yes. Optimised models run on devices such as NVIDIA Jetson modules, smart cameras and accelerators for real-time processing without sending video to the cloud.
How many images are needed to train a model?
With pre-trained models, some tasks begin with hundreds of labelled images per class. Complex or variable tasks may need thousands.
Next step
Exploring Computer Vision?
Tell us what you want to achieve. We will help you judge whether this technology is the right fit and how to test it with your own data.