Computer Vision
AI branch enabling machines to analyze and interpret images and videos.
AI branch enabling machines to analyze and interpret images and videos.
Définition détaillée
Computer vision is an AI field that trains machines to extract useful information from images, videos, and visual streams. It uses convolutional neural networks (CNNs) and vision transformers (ViT) for tasks like image classification, object detection, semantic segmentation, pose estimation, and facial recognition. Combined with OCR, it enables applications like license plate reading, visual quality control, and complex document analysis.
Cas d'usage
Visual quality control on production line: a camera analyzes each part in real-time, detecting defects (scratches, cracks, discoloration) with accuracy surpassing the human eye.
Termes associés
Questions fréquentes
How many images are needed to train a vision model?
It ranges from a few hundred (fine-tuning a pre-trained model) to several million (full training). Data augmentation techniques reduce this requirement.
Does computer vision work in real-time?
Yes, optimized models like YOLO or MobileNet achieve 30 to 60 frames per second on standard hardware.