Computer vision turns pixels into structured facts: counts, positions, labels, measurements. Modern systems rely on deep learning models trained on labeled images. In IoT it is the sensor of choice when the thing to be measured has no electrical signal, for example whether a shelf is empty, whether a worker wears a helmet, or whether a weld looks right.
Video is heavy, so vision workloads push hard toward the edge: a camera that runs object detection locally and transmits only events saves bandwidth and avoids streaming faces to a remote server. The trade-off is hardware cost, since real-time vision needs an NPU or GPU, and sensitivity to lighting and camera placement. See On-device computer vision for IoT for implementation choices.
Part of the IoT Glossary. Spotted an error or a missing term? Tell us.