AI recognition camera: focuses on helmet/hardhat recognition and cat/dog detection with on-device AI, instant HTTP alarm push, algorithms customizable per scenario.
Same hardware platform (HI3516CV610 + SC4336P), different algorithms and mounting specs
An intelligent network camera for construction sites, factories, mining, power & petrochemical and municipal work zones. Detects whether personnel correctly wear safety helmets or industrial hard hats, unified as worn / not worn / improperly worn.
An intelligent network camera for pet stores, kennels, veterinary clinics, boarding facilities, home care and rescue centres. Separates cat from dog first, then determines the specific breed — 135 breeds in total.
Detection and recognition run on-device; results push over HTTP within seconds to link with sirens and third-party systems
Draw the zones that need recognition — gates, corridors, warehouse corners — and assign different rules per region.
Algorithms run inside the camera to detect and classify targets in each zone, working even when the network is down.
Recognition events push over HTTP within seconds, linkable to sound-light alarms, SMS, broadcasting or third-party systems.
Headgear rules for site compliance, pet rules for storage and care — configured per region
Detects whether personnel correctly wear protective headgear, unified as worn / not worn / improperly worn. Safety helmets for sites and factories; hard hats for power, petrochemical, fire and mining.
Detects cat and dog targets, separates cat from dog, and supports multiple simultaneous targets, each labelled with box, class and confidence.
Determines the specific breed on top of the cat / dog class, 135 breeds in total. Out-of-list breeds fall back to the closest known class; only the class is output when confidence is low.
Detects and extracts face regions to confirm whether people are present. No identity matching — it answers "is someone there", not "who is it".
Detects personnel targets to filter false alarms. In helmet scenarios, human detection runs before headgear judgement, cutting misdetections in crowded scenes.
Links with the platform and sends alarms on detection or anomaly; captures and uploads an image to a specified HTTP address (JPG) on event, and can push on a timed schedule.
Recognition capabilities and hardware configuration — exact model parameters confirmed by sampling
| Item | Specification |
|---|---|
| Core algorithm | Headgear recognition: safety helmet and hard hat merged into one, unified as worn / not worn / improperly worn |
| Algorithm output | Detection box + headgear class label, triggerable alarm |
| Assist capability | Face region detection + snapshot; human detection filters false alarms |
| Mounting height | 2.5m ~ 4m, tilt 15° ~ 30° |
| Recognition distance | 3m ~ 15m from lens, detection box at least 1/8 of frame height |
| Framing | Head area fully visible; avoid facing strong light directly |
| Item | Specification |
|---|---|
| Core algorithm | Cat / dog recognition: separates cat from dog first, then determines the specific breed |
| Breed coverage | 135 breeds (12 cat + 123 dog) |
| Algorithm output | Box + cat / dog label + breed name + confidence |
| Assist capability | Face region detection + snapshot; human detection filters false alarms |
| Mounting height | 1.5m ~ 3m, tilt 10° ~ 25° (lower than typical personnel monitoring) |
| Recognition distance | 1m ~ 10m from lens, target area at least 1/10 of frame height |
| Framing | Full body or head visible; dark-coated pets need more illumination |
| Item | Specification |
|---|---|
| SoC | HiSilicon HI3516CV610 (paired with SC4336P) |
| Sensor | SC4336P, 1/3" CMOS |
| Resolution | 4 MP (2560 × 1440) |
| Resolution options | 2560×1440 / 1920×1080 / 1280×720 @30fps (sub 640×480) |
| Lens | 2.8mm ~ 6mm optional, other specs by order |
| Video encoding | H.264 / H.265, 32 Kbps ~ 40 Mbps |
| Day / night | ICR infrared-cut, day / night / auto |
| Protocols | ONVIF, GB28181, RTSP, RTMP, Hikvision VMS-4200 supported |
| Network | 1 × 10M/100M auto-negotiation RJ45; WiFi 2.4G / 4G module (mutually exclusive) |
| Power | DC 12V (10.8V ~ 13.2V), 12V / 100mA class |
| Protection | IP66, all-weather outdoor use |
| Operating temp. | -20℃ ~ 80℃, humidity < 95% non-condensing |
| Form factor | Box / bullet models (same hardware, algorithm and interfaces) |
| Alarm output | HTTP POST of snapshots and results, integrable with your own platform |
| Runtime | On-device AI, local recognition, no cloud dependency, works offline |
What it can detect, how to configure rules, and how to wire up alarms
Safety helmets and hard hats are merged into a single headgear capability — the question a customer actually asks is whether people in the zone are wearing the required gear. When not worn or improperly worn, the camera outputs the judgment and raises an alarm directly.
Detects cat and dog targets, separating cat from dog first, then determining the specific breed — 135 breeds (12 cat + 123 dog). Outputs box, class, breed name and confidence, with multiple simultaneous targets supported.
Detects face regions and outputs snapshots to confirm whether people are present. This does not perform identity matching — it answers "is someone there", not "who is it".
Recognition events push over HTTP within seconds; on an event the camera captures and uploads an image to a specified address (JPG), and can also push on a timed schedule.
Algorithms run on the device with no cloud dependency. Two direct benefits: footage need not leave the site, and recognition and alarms keep working when the network drops.
The most common obstacle to deploying recognition cameras is false alarms. Draw region boxes so recognition runs only where a judgment is actually required, and give each region its own rules.
Available as full device or module — final parameters confirmed by sampling
Tell us what to detect, the install scenario and how alarms should be integrated — you'll get a configuration and sampling plan within 1 business day
Request Quote