supports up to 20 real-time video streams for simultaneous object detection and behavior recognition. Integrated with the latest Deepseek Janus-Pro multimodal AI model, the system combines edge computing, video decoding, intelligent analytics, and event alerting in a single compact device.
| Processor | 8-core A53 CPU (Taishan V200M), 2.3GHz; supports FP32/FP16/BF16/INT8 mixed precision |
| GPU / NPU | Integrated GPU + NPU delivering 36 TOPS performance; supports TensorFlow, PyTorch, Caffe, MXNet, PaddlePaddle, ONNX |
| Memory / Storage | 16GB DDR4 + 64GB eMMC + 64GB TF (expandable to 512GB); supports M.2 SSD expansion |
| Video Input | Up to 20 channels (2–4MP H.264/H.265 IP streams) |
| Video Output | 1× HDMI (supports 1/4/8/9/16/20 split-screen) |
| Audio Output | 1× Audio Line Out |
| Network | Dual 10/100/1000 Mbps RJ45 ports; optional 4G/5G module |
| USB Interface | 2× USB 3.0 (backward compatible with USB 2.0) |
| Operating System | Embedded Linux |
| AI Algorithms | Deepseek Janus-Pro multimodal model + YOLO-based recognition |
| Operating Temperature | -20°C ~ +60°C |
| Operating Humidity | 5% – 90% (non-condensing) |
| Power Supply / Consumption | DC 12V / ≤36W |
| Dimensions | 181 × 150 × 36 mm |
| Weight | ≤1.2 kg (including HDD) |
