A ride-share company wants a system that watches live video from parking-lot cameras and immediately alerts security the moment it detects a person entering a restricted loading zone, without any human reviewing the footage first. Which combination of considerations should the team prioritize when choosing a computer vision approach for this specific requirement?
Select an answer to reveal the explanation.
Short Explanation
You need an object detection model that runs in near real time on the live stream, because the alert has to fire the instant someone steps into the restricted zone. Hourly checks, captions for later review, and signage-text reading all miss the point: they're too slow, still need a human, or answer the wrong question.
Full Explanation
The correct choice is the object detection model that runs in near real time on the video stream. The requirement is immediate detection of a person entering a specific zone within a live video feed, which demands locating where a person is within each frame and doing so continuously enough to trigger an alert the moment the event happens; that combination of localization and near real-time processing is what object detection on a video stream provides. The single-label image classification model applied once per hour to a sampled frame is incorrect because sampling only one frame per hour would miss the vast majority of the time window during which a person could enter and exit the zone, directly contradicting the requirement for an immediate alert. The image captioning model that generates a text description of each frame for a security officer to read later is incorrect because generating a text caption for a human to read later introduces a delay and a manual review step, which the scenario explicitly says should not happen. The model that reads signage in the loading zone using optical character recognition is incorrect because reading signage addresses whether the zone is labeled correctly, not whether a person has entered it, so it does not fulfill the detection-and-alert requirement at all.