Model reference — Live feed
This service detects and follows objects on a live camera — vehicles, equipment, people, animals, goods — and raises an alert when something enters a zone, stops where it should not, moves the wrong way or goes missing.
Because the purpose is to act while something is happening, the whole design is governed by latency and by the alert rate. A system that is right ninety-nine times in a hundred but watches a hundred thousand frames an hour will produce hundreds of false alerts a day and will be switched off by the people it was built for. So live deployments spend as much effort on the rules — how long a condition must persist, how many frames confirm it, which zones and times of day apply — as on the models themselves.
Models in this group take a live camera or stream as input and must keep pace with it in real time. Each specification table gives three hardware tiers — Minimum, the smallest setup on which the model runs correctly; Medium, the usual production configuration; and High, a configuration sized for peak volume. The rate is how many camera streams or feeds one server of that tier can keep up with in real time, not a per-hour count: live work must fit inside the interval between frames, and a server that cannot keep pace drops frames rather than falling behind. Use these rates for initial sizing. Before production, benchmark your own data to validate accuracy, latency, throughput and cost. The sample input here is one 1080p stream at 25 frames per second.
Vendor: Ultralytics
What it does: the standard live detector: several sizes, so accuracy can be traded against streams per server. Fitted to your own objects, it underlies most live deployments.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 6 GB | 24 GB | 80 GB |
| vCPUs | 4 | 8 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (camera streams handled at 25 frames per second) | ≈ 10 streams | ≈ 36 streams | ≈ 100 streams |
Vendor: Ultralytics
What it does: the previous generation, supported by more third-party camera and recorder tooling, which often decides integration with existing equipment.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 6 GB | 24 GB | 80 GB |
| vCPUs | 4 | 8 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (camera streams handled at 25 frames per second) | ≈ 12 streams | ≈ 40 streams | ≈ 110 streams |
Vendor: Baidu (PaddlePaddle)
What it does: no duplicate-removal step and therefore steadier timing, which matters when several models share one frame budget.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 8 GB | 24 GB | 80 GB |
| vCPUs | 6 | 12 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (camera streams handled at 25 frames per second) | ≈ 5 streams | ≈ 18 streams | ≈ 50 streams |
Vendor: University of Science and Technology of China
What it does: better on small and overlapping objects, which is where wide-angle live views usually fail.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 8 GB | 24 GB | 80 GB |
| vCPUs | 6 | 12 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (camera streams handled at 25 frames per second) | ≈ 4 streams | ≈ 15 streams | ≈ 42 streams |
Vendor: Huazhong University of Science and Technology
What it does: follows objects between frames so an alert fires once per object rather than once per frame. Cheap and essential.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 6 GB | 24 GB | 80 GB |
| vCPUs | 4 | 8 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (camera streams handled at 25 frames per second) | ≈ 12 streams | ≈ 40 streams | ≈ 110 streams |
Vendor: Tel Aviv University
What it does: re-identifies an object after occlusion, which prevents one event being reported as several in busy scenes.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 6 GB | 24 GB | 80 GB |
| vCPUs | 4 | 8 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (camera streams handled at 25 frames per second) | ≈ 6 streams | ≈ 22 streams | ≈ 60 streams |
Vendor: Tencent AI Lab
What it does: detects objects from a written list without being fitted first, which lets a new alert condition be trialled the same day it is requested.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 8 GB | 24 GB | 80 GB |
| vCPUs | 6 | 12 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (camera streams handled at 25 frames per second) | ≈ 3 streams | ≈ 12 streams | ≈ 34 streams |
Vendor: Meta
What it does: outlines a tracked object precisely where the alert depends on size or area rather than presence, at a considerably higher cost per stream.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3090 | RTX 4090 | A100 80 GB |
| VRAM | 12 GB | 24 GB | 80 GB |
| vCPUs | 8 | 12 | 24 |
| RAM | 32 GB | 48 GB | 96 GB |
| Server | 1× RTX 3090 24 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (camera streams handled at 25 frames per second) | ≈ 1 stream | ≈ 4 streams | ≈ 10 streams |
The right configuration depends on your objects, your camera coverage, and above all on the alert rate your operators can act on. Our consultants review your views and alert conditions, then recommend the models, the confirmation rules and the hardware for your stream count.
At the start of a project we may run a short proof of concept on a sample of your own data, measuring the accuracy and the throughput the model actually achieves on your material. That replaces the estimates on this page with real figures, so the cost and the schedule for the full engagement are known before it is committed.
Share a representative sample, expected volume, latency target and deployment location for Live object detection / tracking. We benchmark the shortlisted models, recommend the lowest-cost GPU configuration that meets the target, and scale it from pilot capacity to a dedicated production cluster.