Object detection
Object detection is a computer vision task in which an AI model finds specific things in an image or video frame, such as people, cars, animals or products, and marks each one with a label and a bounding box (a rectangle showing its position). In video, detection runs across many frames, so software can tell when and where each object appears.
Why it matters
Object detection turns pixels into searchable facts. It powers product tagging, smart cropping, moderation and footage search, all of which save editors and reviewers time. Detection quality directly affects how useful those features feel to viewers and staff.
How it works
- Each sampled frame is fed to a trained model, often a convolutional or transformer neural network.
- The model finds regions where an object might be and predicts a label for each.
- Each detection gets a bounding box and a confidence score.
- In video, detections are linked across frames so the same object can be tracked over time.
What is the difference between object detection and image classification?
Image classification gives one label for a whole picture, such as "contains a dog." Object detection finds each item separately and draws a box around it, so it can report two dogs and a ball and where each one is. Detection is the step needed for tracking, cropping or counting things in video.
What is YOLO in object detection?
YOLO, short for "You Only Look Once," is a family of object detection models that predict boxes and labels for a whole image in a single pass. That design made real-time detection practical on ordinary hardware. Many later versions have been released by different groups, and YOLO remains a common choice for video because of its speed.
Where it's used
Newsroom editors use object detection to find shots of a specific vehicle or landmark. Creator platforms turn detected objects into search tags and product links, and SaaS products crop video around the subject.
By the numbers
- The original YOLO paper, submitted in June 2015, reported that its base model processed images in real time at 45 frames per second. (You Only Look Once: Unified, Real-Time Object Detection (arXiv))
