IP67-rated AI security camera feature Rockchip RV1126B or RK3576/J/M SoC for commercial, industrial, and automotive applications

Back in January 2024, Firefly released the CT36L AI smart security cameras, built around the Rockchip RV1106G2 SoC with a 0.5 TOPS NPU and Power over Ethernet (PoE) support. Now, Firefly has introduced two new AI cameras, the CQ38W-1126B and CQ38W-3576, which use a similar IP67-rated enclosure but come with much more powerful processors.

Both new models no longer support PoE and instead use a 12V DC input, and they also add an RS485 interface. In terms of performance, the CQ38W-1126B is built around the Rockchip RV1126B with a 3 TOPS NPU and can run small multimodal AI models. The higher-end CQ38W-3576 features an octa-core Rockchip RK3576 with a 6 TOPS NPU, making it suitable for more demanding AI workloads, including YOLO and large language models. Both cameras are available with 3MP or 5MP sensors and come in Commercial or Industrial (J-suffix) variants. The 3576 series also adds an Automotive-grade (M-suffix) option.

Firefly CQ38W-1126B and CQ38W-3576 specifications:

• SoC
  • CQ38W-1126B – Rockchip RV1126B or RV1126BJ (industrial-grade version)
    • CPU – Quad-core Arm Cortex-A53 up to 1.6 GHz with 32KB L1 I-Cache and 32KB L1 D-Cache, unified 512KB L2 Cache
    • GPU – 2D Graphics Engine
    • VPU
      • Video Decoder – H.265/H.264 up to 3840×2160 @ 30fps
      • Video Encoder  – H.265, H.264, JPEG up to 12Mbps @ 30fps
      • JPEG Decoder
    • AI accelerator – Rockchip NPU engine up to 3 TOPS (INT8); supports INT4, INT8, INT16, and FP16 models; TensorFlow, ONNX, PyTorch, and Caffe frameworks.
  • CQ38W-3576 – Rockchip RK3576 /RK3576J/ RK3576M (commercial/industrial/automotive version)
    • CPU – Octa-core  CPU with 4x Cortex-A72 cores at 2.2 GHz, 4x Cortex-A53 cores at 2.0 GHz (1.6GHz for Industrial and Automotive)
    • GPU – Arm Mali-G52 MC3 GPU with support for OpenGL ES 1.1, 2.0, and 3.2, OpenCL 2.0, and Vulkan 1.2
    • NPU – 6 TOPS (INT8) AI accelerator with support for INT4, INT8, INT16, BF16, TF32 mixed operations.
    • VPU
      • Video Decoder
        • H.265, VP9, AV1, and AVS2 up to 8Kp30 or 4Kp120
        • H.264/AVC and MJPEG up to 4Kp60
      • Video Encoder – H.264, H.265, MJPEG up to 4Kp60
• Memory
  • 1GB, 2GB, or 4GB LPDDR4/LPDDR4X (CQ38W-1126B only)
  • 4GB, 8GB, or 16GB LPDDR4/LPDDR4X (CQ38W-3576 only)
• Storage
  • 8GB, 16GB, or 64GB eMMC flash (CQ38W-1126B only)
  • 16GB to 256GB eMMC flash (CQ38W-3576 only)
• Camera options
  • 3MP SONY SC3336 1/2.8″ CMOS sensor (2304×1296) with an F2.56 aperture (F2.0 on the 1126B), 3.95mm focal length, and manual focus
  • 5MP SONY IMX335 1/2.8″ CMOS sensor (2592×1944) with an F2.0 aperture (F2.56 on the 1126B), 4mm focal length, and fixed focus
  • All the variants of the smart camera can be configured with 3MP or 5MP cameras
• Networking – 1x 10/100Mbps Ethernet (RJ45)
• Serial – RS485 via 2-pin 3.5 mm Phoenix terminal block
• Power – 12V DC/2A via 5.5×2.1mm barrel jack
• Dimensions – 69.72 x 69.72 x 174.21 mm
• Weight – Approximately 336 grams
• Ingress Protection Rating – IP67 waterproof and dustproof
• Environment – Temperature range: -20°C to 50°C; humidity: 15%~90%RH (non-condensing)

In terms of software, the CQ38W-1126B supports Debian 12 and Linux systems built with Buildroot + Qt GUI, while the CQ38W-3576 is said to support Ubuntu 22.04 and Debian 12. Both cameras are compatible with common AI frameworks for models. The CQ38W-1126B can run lightweight multimodal and LLM models up to ~2B parameters, including the Qwen series, Gemma2-2B, Phi2, and MiniCPM, and supports optimizations such as weight sparsification and W4A16/W8A16 mixed-precision quantization. The higher-end CQ38W-3576 supports deployment of YOLO and larger language models for more demanding AI workloads. Firefly generally provides a wiki for all of its products, but at the time of writing, I can see that the Wiki is available only for the CQ38W-3576 camera.

Both cameras utilize an “AI-ISP” (an 8MP one in the 1126B and a 16MP one in the 3576), where the 1126B pairs this with AI Remosaic technology to allow for adaptive day/night dual-mode operation without wasting the main NPU’s resources. According to the datasheet, the 1126B features 6-DOF stabilization to actively eliminate high-frequency jitter. It’s kind of weird that the higher-tier RK3576 variant does not have these stabilization features. It’s highly likely the RK3576 variants are based on the Firefly CAM-3576 series SBC we covered last December.

Other AI-enabled smart security cameras include Seeed Studio’s Modbus Vision RS485 and SenseCAP A1102 (LoRaWAN) outdoor Edge AI cameras, the Reolink Floodlight 4K, and camera kits like All-in-One Raspberry Pi CM5 AI Camera Kit from Arducam.

The CQ38W-3576 AI camera is listed on the Firefly store for $149.00. Pricing for the CQ38W-1126B has not been announced yet. More details about the CQ38W-1126B and CQ38W-3576 smart AI cameras are available on their respective product pages.

Debashis Das

Debashis Das is a technical content writer and embedded engineer with over five years of experience in the industry. With expertise in Embedded C, PCB Design, and SEO optimization, he effectively blends difficult technical topics with clear communication

Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress. We also use affiliate links in articles to earn commissions if you make a purchase after clicking on those links.