See the full selection guide
Public Space Crowd Analysis: Tackling Occlusion and Missed Detections
Dense crowd flow analysis in European public safety scenarios, powered by NVIDIA Jetson Orin NX, focusing on occlusion and missed-detection challenges.
Intelligent Video Analytics
Anomaly Detection · People & Vehicle Counting · Industrial Inspection — Compute Right at the Camera
Local AI for every camera — real-time inference, privacy-safe. From 1-stream edge boxes to 20+ channel hubs, choose compute & deploy in minutes.
Scenario Overview
Edge AI for every camera — object detection, OCR, anomaly discovery. Local inference, real-time, privacy-safe. 1–20+ streams, offline-capable. Deployable across industrial, retail, security & healthcare.
Local Real-Time Inference
- NPU/GPU acceleration, ms response
- Video stays on-device, privacy compliant
- Runs offline without network
Multi-Stream Scalability
- 1–20+ channels at 1080p per device
- Tiered compute: 26 TOPS to 275 TOPS
- Horizontal scaling, no bottleneck
Bandwidth-Efficient & Privacy-Safe
- Metadata only — tens of KB per event
- On-device redaction keeps data local
- Dual RTSP / WebRTC, flexible output
Application Scenarios
Missing helmet, someone down — alert the moment it happens
On construction sites, shop floors and substations the question is whether people follow the rules; in care homes, single-occupancy housing and hospital corridors it is whether someone is in trouble. Both share one trait — the incident lasts seconds, and reviewing footage afterwards changes nothing. Edge models judge PPE compliance and body posture directly in the frame and fire a local audible/visual alert the instant something goes wrong.
Core Advantages
- Combine helmet, hi-vis vest, gloves and mask checks to match site rules, applied only inside the zones you define
- Falls are judged from skeletal pose, not guessed from "the person stopped moving" — prolonged lying and loitering are flagged too
- Alerts drive local sirens, lights or access control on the device itself, keep working offline, and send back only tens of KB of event metadata
PPE Compliance Detection
Instant alerts for missing helmets or vests, scoped by zone and shift
Fall Detection
Skeletal pose detects falls and prolonged lying — care homes, solo housing, hospital corridors
Restricted Zone Intrusion
Draw no-entry zones and trip lines; unauthorized entry triggers a local alert
How many came, where they lingered, how many vehicles
Stores need footfall and conversion, districts and city operators need hourly crowd load, car parks and gatehouses need vehicles in and out. The camera outputs three metric families at the edge — entry/exit counts, dwell heatmaps and vehicle flow — and sends back numbers only: no bandwidth cost, no faces.
Core Advantages
- Footfall — bi-directional counting with ID de-duplication, so someone pacing back and forth is counted once
- Activity — dwell time per zone aggregated into heatmaps, showing which shelf or which junction runs hottest
- Vehicle flow — in/out counts for car parks, gatehouses and road cross-sections with the same operator set
- Output is counts and durations, ready for BI dashboards — the video never leaves the device
Footfall Counting
Bi-directional counting with ID tracking — store footfall and venue load, straight out
Dwell & Activity Analysis
Dwell time as heatmaps and hourly curves — see the flow lines and the peaks
Vehicle Flow Counting
In/out vehicle counts and cross-section flow — one operator set for car parks and gatehouses
Scratches, missing parts, batch codes — stopped on the line, not after
Human visual inspection drifts over a shift, and the same batch gets judged differently at the start and at the end. Edge vision completes three checks within line takt time — surface defects, missing or wrong parts, and character/batch codes — rejecting NG units on the spot and logging each verdict with station, timestamp and batch, so a problem can be traced to the individual unit.
Core Advantages
- No need to build a defect dataset from scratch — add hundreds to a few thousand on-site images to an open dataset and fine-tune
- Character and QR recognition runs on-device, auto-matching model, serial and production date against the work order
- Verdicts are logged with station, timestamp and batch, supporting batch-level lookback
Surface Defect Inspection
Scratches, dark spots, contamination and wrong models — judged unit by unit within takt time
Missing & Misassembly Detection
Missing screws, reversed components, absent parts — caught before the unit moves on
Character & Batch Code Verification
On-device OCR for printed, laser-etched and QR codes — auto-matched to the work order and written to the traceability log
Deployment & Selection
Where to place compute: 3 deployment topologies
A video analytics solution combines four types of components — cameras, AI compute nodes, network, and cloud. Depending on where compute resides, deployments fall into three topologies.
Core Advantages
- All-in-One AI Camera (reCamera / R2130): capture and inference in one device, 1 stream, plug-and-play
- Detached Edge Box (J3011 / J4012 / J5012): reuse existing IP cameras over RTSP — the path for anything from 2 to a dozen-plus channels
- Camera-Direct Compute Unit (J5012 + GMSL): cameras wired straight to the compute unit over coax instead of Ethernet — automotive-grade link for robotics, AGVs and in-vehicle use
| Product Image | Product | Core Positioning | AI Performance (INT8) | Channels (YOLOv8m) | Cost |
|---|---|---|---|---|---|
 | Vision V2 | OEM Custom Board | 0.04 TOPS | — (tiny models only) | — |
 | reCamera | All-in-One AI Camera | 1 TOPS | 1 ch | ~$80 |
 | reCamera Pro | All-in-One AI Camera (4K + VLM) | 3 TOPS | 1 ch | ~$300 |
 | R2130 | RPi 5 + Hailo-8 | 26 TOPS | 2–3 ch | ~$370 |
 | J3011 | Jetson Orin Nano 8G | 40 TOPS | ~4 ch | ~$750 |
 | J4012 | Jetson Orin NX 16G | 100 TOPS | ~7 ch | ~$1400 |
 | J5012 | AGX Orin 64G | 275 TOPS | ~15 ch | ~$4550 |
Right-sizing: choose hardware by channel count
To determine how many 1080p30 streams you can run, look at TOPS and model size. The table below uses **YOLOv8m** (industrial-grade accuracy) as the common baseline, summarizing AI performance, channel count, and cost across seven products. Baseline details: 1080p @ 15 FPS INT8 (640×640); for 30+ FPS scenes, halve the estimate.
Core Advantages
- Channel benchmark: estimated on the mainstream industrial model (YOLOv8m); lighter models run 2–3× more — see table notes
- Bottleneck: video decode, not inference — larger models amplify; real-world figures run slightly below pure-inference estimates
- Prices are whole-unit reference figures at the time of lookup, for order-of-magnitude only; they vary with configuration and volume — the product page governs
- How to choose: 1 ch — reCamera, or reCamera Pro when you need 4K or VLM; 2–6 ch — J3011/J4012; 8+ ch, or GMSL direct-connect cameras — J5012
| Product Image | Product | Ethernet | USB | MIPI CSI | GMSL | RTSP Decode (1080p30, H.265) |
|---|---|---|---|---|---|---|
| reCamera | ×1 (100M) | ×1 | — | — | 1 stream (on-device) |
 | reCamera 2002 HQ PoE | ×1 (100M, PoE-powered) | ×1 | — | — | 1 stream (on-device) |
| reCamera Pro | ×1 (GbE) | ×1 (Type-C 3.0) | ×2 (1 used by built-in 8MP sensor) | — | 1 stream (on-device, 4K30 decode) |
| R2130 | ×1 (GbE) | ×2 | ×2 | — | ~4 streams (RPi 5 VPU) |
| J3011 | ×1 (GbE) | ×4 | ×2 | — | 11 streams |
| J4012 | ×1 (GbE) | ×4 | ×2 | — | 18 streams |
| J5012 | ×4 GbE + ×1 10GbE | ×4 | — | ×8 | 44 streams |
Which device works with my camera?
The table below summarizes video input interfaces and hardware decoding limits for each device.
Core Advantages
- IP Cameras: RTSP over Ethernet; channels capped by HW decode (NVIDIA figures)
- Local Cameras: MIPI CSI (J3011 / J4012 / reCamera Pro) or GMSL (J5012 only, ×8)
- Output & Bus: RTSP / WebRTC out; CAN on J4012 / J5012 / reCamera Pro, RS-485 on J4012 / J5012
- All-in-One Cameras: 2002w — USB-C + WiFi, indoor plug-and-play; HQ PoE — one cable for power + data, IP66, M12 lenses; Pro — 4K + VLM over GbE
| Product Image | Product | Cooling Method | Suitable Environment | Operating Temperature | Power Supply |
|---|---|---|---|---|---|
 | reComputer AI Industrial R2135 | Active (Fan) | Standard Indoor | -20~65°C | DC 12-19V |
 | reComputer Industrial R2235 | Passive (Fanless) | Dusty/Workshop | -20~50°C | DC 9-36V Wide |
 | reComputer Super J3011 | Active (Fan) Super | Standard Indoor | -20~60°C | DC 12-19V |
 | reComputer Industrial J3011 | Passive (Fanless) Industrial | Dusty/Workshop | -20~60°C | DC 12V~24V |
 | reComputer Super J4012 | Active (Fan) Super | Standard Indoor | -20~60°C | DC 12-19V |
 | reComputer Industrial J4012 | Passive (Fanless) Industrial | Dusty/Workshop | -20~60°C | DC 12V~24V |
| reComputer Robotics J5012 | Active (Fan) | Standard Indoor / GMSL | -10~60°C | DC 19~48V |
 | reServer Industrial J501 | Passive (Fanless) Industrial | Dusty/Workshop | -20~60°C | DC 12V~36V |
Passive vs Active: choose by dust level
Dust level determines cooling: fan (active) for lower cost & sustained full-load; fanless sealed enclosure (passive) for dust, wide temp & voltage — built for harsh environments.
Core Advantages
- High dust (mining, mills, outdoor) → passive Industrial/Fanless, sealed fanless
- Indoor (office, retail, lab) → active cooling (Super/AI), fan-cooled better value
- Naming: J-series 'Industrial' = passive; R2000 'AI' = fan, else fanless
Success Cases
FAQ
Contact Us
We Are Glad to Be Your Hardware Partner !