A smart-city pilot discusses edge AI cameras that process locally to reduce raw video leaving intersections. What edge AI concept applies?

Edge AI means running inference on or near the municipal device—such as an intersection camera—so raw data need not all be shipped to a central cloud for every analytic. A smart-city pilot that processes locally to reduce raw video leaving the intersection is applying that concept for bandwidth, latency, and privacy-aware sensing. Edge processing can emit events, counts, or redacted outputs instead of continuous raw streams, which is often desirable in civic deployments.

Off-site cloud model inference that can limit how much raw municipal video must leave the intersection is not the edge AI concept here, because inference off-site requires sending raw video away for analysis rather than running inference on or near the device to produce events or counts. Central data-center model training that can limit how much raw municipal video must leave the intersection is not edge AI, because training builds model weights centrally rather than inferring on or near the device. Curb-side model quantization and pruning that can limit how much raw municipal video must leave the intersection is not the edge AI concept here, because quantization and pruning only optimize a model to fit constrained hardware rather than running local inference to limit raw streams.

Exam caveat: edge AI does not automatically equal privacy compliance—local models can still collect sensitive imagery, and derived data may remain personal data under policy. Operational check: document what leaves the intersection, set retention for both raw and derived data, secure device update channels, and verify that analytics purpose limitation matches the pilot’s public notice.