Technology that enables computers to interpret and understand images and video, detecting objects, scenes and visual patterns.
Also known as: CV, image recognition, visual AI
In plain English
Computer vision lets machines 'see' and make sense of photos or video, like a phone recognising faces or a shopping app finding similar items. It combines cameras and software to label, count or track things in images. Many everyday features—auto‑tagging, augmented reality and visual search, use it without you needing to know the technical details.
In practice
In practice, computer vision appears as automated inspection, image moderation, visual search or analytics embedded in products and workflows. Procurement focuses on accuracy on target data, latency for real‑time use, privacy for video feeds and ease of integration into pipelines. Good vision models reduce manual review, speed decisions and lower operational risk when deployed correctly.
Under the hood
Computer vision systems typically use convolutional neural networks or vision transformers to extract hierarchical visual features from preprocessed images, trained on labelled datasets with data augmentation and transfer learning. Key engineering concerns include dataset curation, loss functions, evaluation metrics (e.g. mAP, IoU), inference latency, and deployment on GPUs or edge accelerators. Modern systems often combine pretrained foundation models and fine‑tuning for domain‑specific performance.
Example
Can we apply computer vision to automatically detect defects on the assembly line and feed results into our monitoring dashboard?
Related terms
Deep Learning · Neural Network · Machine Learning · Multimodal AI · Foundation Model