Remote projects often begin with an inconvenient fact: there is no NVR to integrate, the camera owner cannot provide an RTSP stream, broadband is weak, or the point is connected only through a mobile network. A traditional edge box can be a good fit for an existing multi-camera LAN. It is not always the first choice for a single, scattered point with uncertain video access.
An AI camera changes the starting point. The camera captures the image and runs the selected detection locally. It can create a snapshot, event time, device name, rule result, and optional short evidence clip. A 4G or other network link then carries the event information to the office or platform when the link is available. Continuous transmission of every camera frame is not required for local detection.
Questions to answer before selecting the device
What does the camera need to detect?
Start with one or two high-value events per point, such as PPE, smoke and flame, vehicle entry, a restricted-zone rule, or a site-specific object. The target has to be clear enough in the camera view. A broad list of algorithms is less useful than a defined event that the local team will actually review.
How will the site connect?
Test mobile coverage at the planned mounting location. Signal bars on a phone are a rough indication only. The project should check uplink stability, data plan limits, power supply, weather exposure, remote access requirements, and how much evidence is sent per event. If the network fails, local detection and local event recording can continue according to the deployed configuration. The office will not receive a new event until communication returns.
What is kept locally and what is sent centrally?
The design should name the fields sent to the office: image, short clip, time, site, camera, event type, and any approved metadata. It should also state whether original video is retained locally, how long it is stored, whether missed uploads are retried, and who may view the event evidence.
When a hybrid design makes sense
Some customers have a mix of isolated cameras and sites with existing NVRs. They can use AI cameras at the isolated points while adding an AI edge box to a local multi-camera LAN where streams are available. The central office then receives events from both types of device. This design has to account for local network conditions. It should not assume every camera can reach the public internet or that every original stream needs to be centralized.