Object detection answers two questions per object: what is it, and where is it. Unlike plain image classification, which gives one label per frame, a detector outputs a list of boxes with classes and confidence scores. That makes it the workhorse of computer vision in IoT: counting vehicles in a lane, finding packages on a conveyor, spotting a person in a restricted zone.
Detectors are heavier than classifiers, so running them at camera frame rates on an edge device usually means a compact architecture, a lower input resolution and an NPU. Accuracy drops for small or overlapping objects, and a model trained in daylight may fail at night. The guide On-device computer vision for IoT covers model choice and deployment.
Part of the IoT Glossary. Spotted an error or a missing term? Tell us.