Firefly dual-lens AI camera module comes with Rockchip RV1109 or RV1126 processor

We’ve very recently written about Rockchip RV1126 quad-core Cortex-A7 processor designed AI cameras thanks to a 2.0 TOPS AI accelerator together with its little dual-core brother: RV1109.

We’ve also found some RV1109 and RV1126 evaluation boards, but prices were in the hundreds of dollars. It was then brought to our attention than the upcoming MAIX-III board would feature RV1126 at a more affordable cost, but it turns out there’s no need to wait as Firefly has just launched a dual-lens AI camera module based on either Rockchip RV1109 or RV1126 for under $100 with respectively CAM-C1109S2U and CAM-C1126S2U models.

RV1126 dual-lens AI camera moduleBoth version of the modules have the same specifications except for the processor:

  • SoC (one or the other)
    • RV1109 SoC with
      • CPU – Dual-core Arm Cortex-A7 @ 1.5 GHz + RISC-V MCU @ 400 MHz
      • GPU – 2D graphics engine
      • NPU – 1.2 TOPS with support for INT8/ INT16
      • VPU
        • Encoder – H.264/H.265 up to 2,688 x 1,520 @ 30 fps + 1,280 x 720 @ 30 fps
        • Decoder – 5M H.264/H.265
    • RV1126 SoC with
      • CPU – Quad-core Arm Cortex-A7 @ 1.5 GHz + RISC-V MCU @ 400 MHz
      • GPU – 2D graphics engine
      • NPU – 2.0 TOPS with support for INT8/ INT16
      • VPU
        • Encoder – 4K H.264/H.265 up to 3,840 x 2,160 @ 30 fps + 720p @ 30 fps
        • Decoder – 4K H.264/H.265 up to 2x 3,840 x 2,160 @ 30 fps
  • System Memory – 1 or 2GB DDR4 RAM
  • Storage – 8 or 16 GB eMMC 4.51 flash
  • Cameras
    • RGB camera – 1920×1080 resolution GC2093 image sensor with 650nm filter, 80cm focus distance, F2.0/4.3mm
    • IR camera – 1920×1080 resolution GC2053 image sensor with 850nm filter, 80cm focus distance, F2.0/4.3mm
    • Raw output format
  • Video Output – 30-pin FPC header to connect MIPI display (TBC)
  • Power Supply – 5V via USB Type-C port
  • Dimensions – 84 x 22.45 x 19.35 mm (Aluminum alloy housing)
  • Temperature Range – -10 to 60°C

RV1126 Face recognition

The camera module runs Linux, and it supported by Rockchip RKNN toolkit working in Windows, Linux (64-bit x86 and Arm), and Mac OS. The AI camera module connects to a host platform such as an Android tablet. Considering the cameras are all fixed focus with a 80 cm focus distance, the main application is face recognition and detection. There’s no documentation in English for now, but the Chinese version of the Wiki has plenty of information and resources to get started.

The company says the solution supports up to 100,000 faces in the database, although 10,000 is recommended, and offers 99% accuracy without mask, 95% with mask. Performance wise the camera module can typically detect a face within 23ms, human recognition can be done within 15ms using dual-lens detection, and face recognition can be as fast as 0.0115ms. The recommended input image size is 1280×720, but the system works with images as small as 90×90 for human face recognition.

CAM-C1109S2U and CAM-C1126S2U dual-lens AI camera modules are available now for respectively $89 ad $95 on Firefly store. More details may also be found on the product page.

Share this:

Support CNX Software! Donate via PayPal or cryptocurrencies, become a Patron on Patreon, or buy review samples

Subscribe
Notify of
guest
The comment form collects your name, email and content to allow us keep track of the comments placed on the website. Please read and accept our website Terms and Privacy Policy to post a comment.
5 Comments
oldest
newest
Rogan Dawes
Rogan Dawes
8 months ago

I still reckon that people would pay good money for a camera that simply did background elimination, replacing it with a green screen. Given the dual cameras, it should be possible to do depth detection, and show only what is in the foreground. i.e. whatever is within a particular depth range.

Jon Smirl
8 months ago

Get one of these — dual 4K camera and hardware depth mapping. $95 plus image sensors.
http://www.lindeni.org/lindenis-v5.html

You can probably do real-time background removal/replacement completely in hardware by combining the depth map with ISP processing. So LIVE TV! with fake background.

Their image sensors $47ea, but they are extremely good sensors. If you want to make the effort to build a board, the V5 supports several cheaper choices.

Keith Woodward
8 months ago

Thanks for the link Jon and question Rowan. May I ask what specific cameras you pair with that board, Jon?

Jon Smirl
8 months ago

Click on the link for their Aliexpress shop. All of the needed peripherals are for sale there. I bought two of their camera modules. They are good 4K sensors that work with very little light. A bare IMX317 sensor chip is $23. They put it on a PCB and give you good glass lenses with it. The code for the board is up on github. Checkout the Lindeni wiki. Look back in CNX archives, there have been articles about this board. The drawback is the AI unit. The AI unit is based on HAAR instead of CNN. HAAR does work,… Read more »

Keith Woodward
8 months ago

Thanks very much, great advice as always

Advertisement