Volume 1: Quick Start
Estimated time: 10–15 minutes Goal: Go from unboxing to seeing live AI detection results in your browser Prerequisites: You have received a CosmoEdge-preinstalled edge device No extra equipment needed: Built-in demo videos are provided — no camera required to complete the entire tutorial
Hardware Overview
CosmoEdge runs on a Sophon-based AI edge computing device powered by the BM1688 processor, delivering 16.0 TOPS of AI compute for intelligent analysis of IP camera video streams.
Key specifications:
| Component | Specification |
|---|---|
| Processor | BM1688 |
| CPU | Octa-core ARM A53 @ 1.6 GHz |
| Memory | LPDDR4 8 GB |
| Storage | 64 GB |
| AI Compute | 16.0 TOPS |
| Network | 2 × 10/100/1000 Mbps auto-negotiation Ethernet |
| Interfaces | USB 3.0 × 2, Type-C × 1, HDMI × 1, TF × 1, SIM × 1 |
Step 1: Hardware Connection
Connect the device to your local network and power it on.
- Use an Ethernet cable to connect the device's ETH port to your router or switch.
- Plug in the power adapter.
- Wait approximately 60 seconds for the device to finish booting.
Expected status indicators:
- Power LED (PWR): Solid on (red)
- Network LED (WAN): Blinking (indicates a healthy network connection)
Configure a static IP on your computer
On your computer, go to Start → Control Panel → Network and Internet → Network and Sharing Center → Change adapter settings → Ethernet. (A direct connection between the device and your computer is recommended for initial setup.)
- Double-click the Ethernet adapter and open Internet Protocol Version 4 (TCP/IPv4).
- Set the IP address and subnet mask to be on the same subnet as the CosmoEdge device, then click OK.
💡 The default CosmoEdge address is: WAN: 192.168.100.1
Step 2: Access the Management Console and Configure System Settings
Open a browser and navigate to the device's IP address.
- We recommend Chrome or Edge.
- In the address bar, enter:
http://<device IP>(e.g.,http://192.168.100.1)
- Log in with the default credentials:
- Username:
admin - Password:
admin(we strongly recommend changing this after first login)
- Username:
After a successful login, you'll see the System Dashboard, which displays the following key metrics:
| Metric | Description |
|---|---|
| CPU Usage | Processor load |
| VRAM Usage | Video memory consumed by loaded tasks |
| NPU Usage | Neural processing unit load |
| eMMC Usage | System storage utilization |
| Packet Loss | If above 10%, system performance may degrade |
Go to System Management → System Settings → Time Settings → Manual Sync → Sync with Computer. (Since the device is directly connected, it can't automatically obtain the correct time.)
Go to System Management → Network Configuration to change the device's IP address so it's on the same subnet as your LAN. (This lets your computer access the console while still having internet connectivity.)
⚠️ Can't access the management console?
- Make sure the device and your computer are on the same subnet.
- Try connecting the device directly to your computer with an Ethernet cable (bypassing the router).
- Check whether your browser is using a proxy — disable it and try again.
Step 3: View Live AI Detection Results
The device comes pre-loaded with several demo scenarios using built-in demo videos as data sources — no cameras needed to experience the full AI detection workflow. We'll use Pedestrian Flow Counting as our example.
Download the pedestrian flow demo video
Demo video download link
Upload the pedestrian flow demo video
In the Task Configuration menu, click Video Sources, then click Add.
In the dialog that appears:
- Source type: Offline Video
- Channel name: Give it a descriptive name, e.g., Building 1 East Corridor
- Upload video: Upload the demo MP4 file, then click Save.
Click Scenario Task Assignment.
Assign an algorithm to the channel
Configure the video analytics algorithm for the channel.
Page layout overview:
The Service Assignment page has three main areas:
- Area 1: All available algorithm types — Detection/Analysis, Face/Body, Counting/Statistics.
- Area 2: Controls for starting, stopping, and deleting services on the channel.
- Area 3: Service configuration panel — detection region setup, parameter tuning, runtime strategy, etc.
Under Counting/Statistics, find Pedestrian Flow Counting and select it.
The configuration parameters for the selected algorithm will appear. The pedestrian flow counting algorithm works by detecting pedestrians crossing a detection line. Key controls include:
- Draw: Draw a detection line.
- Direction Toggle: Switch the crossing direction for one-sided counting.
- Delete: Remove the detection line.
💡 Configuration Parameters
Parameters are determined by the algorithm's pipeline orchestration.
Draw the detection line
Click Draw, then click once in the video to set the starting point, and click again to set the endpoint. (The line counts pedestrians crossing in the direction of the arrow.)
Click Finish Drawing to save. The detection line is now configured.
Set the runtime strategy
Set the play count to 0 for looping playback.
Click Save. This assigns the pedestrian flow counting algorithm to the channel and simultaneously starts the analysis service.
Return to the Video Sources menu. The "Running" toggle being on indicates the service is active.
- Algorithm Visualization
Click Live Preview to enter the visualization page.
The visualization page has three areas:
- Left panel: All service channels in the system.
- Center area: Visual algorithm display — select a channel to view.
- Right panel: Real-time scrolling alarm feed from all active algorithm tasks.
💡 Layout Options Use the toggle in the upper-right corner to switch between single view and quad view layouts.
Select the Building 1 East Corridor channel to stream its algorithm-enhanced video.
Under algorithm overlay, select Pedestrian Flow Counting. Different algorithms have different overlay behaviors.
💡 Overlay Logic
Each algorithm's pipeline strategy determines its own overlay behavior.
Overlay results:
The visualization includes:
- Pedestrian detection: Object class and confidence score, e.g.,
Pedestrian: 0.91 - Pedestrian tracking: Unique track ID for each person, e.g.,
#1318 - Pipeline timing: Upper-left area shows per-component latency breakdown.
- Counting results: Upper-right area shows IN/OUT pedestrian flow totals.
Demo walkthrough
Video demo: Coming soon.
Step 4: View Statistics
All AI-detected alarms and statistical events are automatically recorded and can be reviewed later.
- In the left navigation, click Event Center → Counting/Statistics.
- Select Query Events → Channel Name → Algorithm Service → click Query. (Click Query again to refresh with the latest real-time statistics.)
Page note: "Departing visitors" = the OUT count from the visualization; "Net inflow" = the IN count.
🎉 Quick Start Complete
You've now successfully:
- [X] Powered on the device and connected it to the network
- [X] Accessed the management console via browser
- [X] Viewed live AI detection results from the pre-installed demo
- [X] Reviewed historical alarm records
What's next:
| Goal | Read |
|---|---|
| Configure your own AI detection scenarios | → Scenario Configuration Guide (Volume 2) |
| Try the VLM — switch detection rules without model training | → VLM Visual State Judgment Guide (Volume 3) |
| Look up a specific parameter or troubleshoot an issue | → Reference Manual |