
Step 1: Select a Camera
The camera is the eyes of your vision system. You can use existing cameras or select new ones based on your scenario's requirements.
Selection Criteria | Description |
Device Type | Standard IP Camera: Suitable for existing surveillance systems, accessible via RTSP protocol. Edge AI Camera: Features built-in computing power to run simple models directly on the device, ideal for scenarios with strict latency and bandwidth requirements. |
Deployment Environment | Choose devices with the appropriate Ingress Protection (IP) rating (e.g., IP66) and power supply method (PoE, DC power) for indoor or outdoor use. |
Communication Method | Supports various connections like Ethernet, Wi-Fi, and 4G to ensure stable video stream transmission. |
Image Quality | Select the appropriate resolution, focal length, and frame rate based on the size and distance of the objects to be identified, ensuring the model receives clear image data. |
Step 2: Select an AI Computing Device
The AI computing device is the brain of the solution, responsible for running vision models for inference. If you are using Standard IP Cameras, a device from this section is required. If you are using an Edge AI Camera, it already has built-in processing, so a separate device may not be necessary for simpler tasks. Choosing the right external computing device is key to balancing cost and performance.
Selection Criteria | Description |
Computing Power (TOPS) | TOPS (Trillions of Operations Per Second) is the core metric for AI chip performance. Higher TOPS enable more complex models and faster processing. (Note: The TOPS listed in this document are dense TOPS, commonly used for vision inference.) |
Performance Requirements | Evaluate the total required computing power based on the number of camera streams to be processed simultaneously and the inference frame rate (FPS) required per stream. For example, processing four 1080p video streams at 30 FPS each requires significantly more power than one 720p stream at 10 FPS. |
Model Complexity | Complex models (e.g., YOLOv8-x) require more computing power to achieve the same frame rate as simpler models (e.g., YOLOv8-n). Use the tables below to match your application needs with our device performance. |
Structural Requirements | With Fan: Suitable for indoor environments with a high demand for a compact form factor. Fanless: Larger in size but offers superior dust and water resistance, making it ideal for outdoor use. |
Task Types & Frame Rates (FPS)
| Near Real-time | Standard Real-time | High-Speed Real-time | Special Applications |
FPS Requirement | <15 FPS | 15 - 30 FPS | 30 - 60 FPS | > 120 FPS |
Application Scenario | Smart Agriculture | Urban Security | Autonomous Driving | Crash Testing |
Note: 24 FPS is the standard for the motion picture industry.
Seeed Studio Edge AI Device Performance Reference (INT8, 640x640)
| reComputer R2130 | reComputer J3011 | reComputer J4012 | Jetson AGX Orin™ 64GB |
fps@yolov8-n | 80-120 | ~90 | ~250 | ~380 |
fps@yolov8-s | 60-90 | ~60 | ~190 | ~300 |
fps@yolov8-m | 30-50 | ~40 | ~90 | ~140 |
fps@yolov8-l | - | ~20 | ~70 | ~110 |
fps@yolov8-x | - | ~15 | ~45 | ~75 |
*Data is for general estimation; actual performance may vary depending on the specific application. For reference only.
Step 3: Select an AI Model Training & Deployment Platform
The SenseCraft AI provides a full-stack toolchain from data annotation and model training to one-click deployment, significantly lowering the barrier to AI development.
Platform Advantage | Description |
No-Code Training | Train and evaluate custom models by uploading datasets through a simple interface, no coding required. |
Rich Pre-trained Models | Access a wide range of pre-trained models for various scenarios, which can be deployed directly or fine-tuned to accelerate development. |
One-Click Deployment | Deploy trained models to designated edge devices with a single click, ensuring seamless hardware-software integration. |
Open Compatibility | The platform supports importing models trained with other mainstream frameworks like TensorFlow and PyTorch for easy integration. |
Step 4: Select a Data & Application Platform
This platform aggregates AI analysis results for storage, visualization, and subsequent automated actions (e.g., alerts, logging).
Deployment Mode | Advantages | Requirements |
SenseCraft Public Cloud (Recommended) | Simple, fast, and maintenance-free. Provides standard APIs and a web interface for quick validation and secondary development. | Edge computing device requires internet access. |
Private On-Premise Deployment | Data remains completely private and can be deeply integrated with internal enterprise systems, meeting high security and customization needs. | Requires self-hosting and maintenance of servers, along with certain technical expertise. |
SenseCraft offers a comprehensive suite of hardware-agnostic tools for everything from device configuration and data retrieval to AI model deployment, enabling you to build your own custom solutions.
Leveraging advanced edge computing and AI visual analysis, our solution processes video streams from cameras in real-time to accurately identify specific objects, behaviors, or anomalies. Whether for production line quality control, public area security monitoring, or smart agriculture crop analysis, this solution provides a full suite of tools and support, from hardware selection and model deployment to application development. With the SenseCraft AI platform, you can effortlessly train and deploy custom models, rapidly integrating intelligent vision capabilities into your business to enhance efficiency and security.