S3-OLinuXino OSHW IP Camera update: board routing complete, next to-do prototype.


Screenshot from 2019-12-03 08-44-18Screenshot from 2019-12-03 08-45-48

S3-OLinuXino is now completely routed and we make prototypes to verify everything is well designed. If the prototypes are OK production will follow in Q1 2020.

22 Comments (+add yours?)

  1. et9000
    Dec 03, 2019 @ 10:28:24

    Nice! It looks like it supports 2 cameras, is that correct?

    Reply

    • OLIMEX Ltd
      Dec 03, 2019 @ 11:28:43

      It has MIPI and CSI connectors, but Jon wrote on the previous post that the software driver will not handle both at same time

      Reply

    • jonsmirl
      Dec 03, 2019 @ 17:00:37

      This is not a good chip to target for dual camera support. At best you might be able to just get it working for dual 1080P30. But… you will probably need a heat sink on it since that is right at the edge of what the chip can do. And it is missing a whole lot of software.

      Use a Lindeni V5 — dual 4K cameras work out-of-the-box. For open source V5 support, the h.264 encode/decode is the same generic Allwinner IP so the bootlin drivers should work after they have been adjusted for pipeline differences.

      Reply

    • jonsmirl
      Dec 03, 2019 @ 17:08:16

      If you just want two image sensors with only one h.264 stream, that is probably more doable. You should be able to switch between the sensors. But.. the MIPI interface is missing an open source driver. bootlin is looking for intern to help write it.

      If the ISP is working (also missing an open source driver) picture-in-picture should be doable. The ISP would scale the second sensor capture and place it into the first sensor capture before encoding.

      One thing you are not considering is the limit of S3 memory bus bandwidth. Dual image streams may exceed the available bandwidth. I have never played with dual stream but I have run into the bandwidth limit on occasion. V5 is designed to have plenty of bandwidth.

      Reply

  2. Lucio De Re
    Dec 03, 2019 @ 16:43:50

    But that is a surmountable obstacle, surely?

    Reply

  3. jonsmirl
    Dec 03, 2019 @ 17:14:00

    What are thoughts on array mic solution? On previous thread I described four options. It is not hard to get Alexa or Baidu working on S3 with mono mic/speaker. As soon as people build the mono solution they immediate ask for an array mic solution since the mono only works in two foot radius. This can be add-on board.

    Reply

    • OLIMEX Ltd
      Dec 03, 2019 @ 17:17:50

      this is just first prototype, array mic could be add via UEXT connector

      Reply

      • jonsmirl
        Dec 03, 2019 @ 17:24:24

        This AC108 based board has the right hardware if you jumper it in. The device driver needs to be ported but that should not be hard. https://respeaker.io/4_mic_array/

        We are using the Knowles IA8201 solution.

      • jonsmirl
        Dec 03, 2019 @ 17:30:48

        Note that the AC108 works in two modes — one I2S line and TDM for four mics, or two I2S lines and no TDS. The respeaker board only supports TDM mode. If you try to reuse the board on non-Allwinner CPUs you may need the non-TDM solution.

        Four mics is enough for 2D. You can use two of the Respeaker boards to build mic cube which will locate in 3D.

  4. jonsmirl
    Dec 03, 2019 @ 17:18:49

    Please use a decent quality TF Card socket with spring ejector, that socket gets a lot of use during software development and it is annoying if it doesn’t work well.

    Reply

  5. xemilo
    Dec 03, 2019 @ 23:11:34

    What is the black 8-pin chip near the “Flash1” header labeled “U2” ?

    1. SPI flash memory – why not, footprint and price is so small, but so much better to have it.
    2. Low profile/half heigth RJ45/LAN/Ethernet port like this one (first result on google):
    https://www.te.com/global-en/product-1888250-1.html

    This will reduce board height and enclosure size so much. Bonus: it is more solid 🙂

    Mailine linux support for S3 is good progress, for V5 is not even started ?

    Reply

  6. Yevgeny
    Dec 04, 2019 @ 17:23:02

    In the datasheet for S3 there is a mention of the presence of an RGMII interface, as well as that the MAC supports gigabit mode. Can anyone here confirm that it is possible to get S3 to work with gigabit PHY? If I correctly understood from your question in a previous post about POE with gigabit speed, then there really are no ready-made solutions with a price of $ 4. The H6062NL transformer costs $ 2, but that’s not all, although I haven’t done the calculation yet.

    My application needs grayscale YUV stream from a parallel camera at least 20 frames per second, which is 82.944 megabytes per second. In this case, there can be more than two cameras in the system, and the physical distance between the cameras is several meters. Gigabit ethernet seems to be a suitable solution at first glance for my task, and SIP S3 is quite budget-friendly. I know about V5 and many other solutions, but I’m looking for SIP and don’t need special processing on this CPU, I just need to stream it. Allwinner has FEL usb mode + NFS, which allows you to use sip without SPI flash at all, if you download an image from the network at every start, but I have not tried this.

    As far as I understand, you do not plan to output DCMI on your stm32mp1 board, in this case PanGu stm32mp1 board has more advantages.

    Reply

    • OLIMEX Ltd
      Dec 04, 2019 @ 17:56:08

      advantage for one may be disadvantage for another 😉 S3-OLinuXino serves our own need for IP camera running Linux, it’s initial proof of concept which we will use as brick inside few other products, SMT32MP1 is build to be connector compatible with LIME2 for industrial customers. LIME2 has camera interface on GPIO headers and same will be for STM32MP1 board we work on but there will be no particular camera installed on it

      Reply

    • jonsmirl
      Dec 04, 2019 @ 19:56:58

      I would use analog output camera (AHD) for this application. Run the coax several meters back to a central SOC that digitizes the frames. You can buy analog output cameras for $10 on Aliexpress. Many of the Allwinner SOCs support analog camera input.

      Reply

      • Yevgeny
        Dec 04, 2019 @ 21:10:01

        One of the proposed modes of operation in my solution is stereo vision. Sensors with a global shutter must be synchronized and even lossy compression is undesirable, so an analog signal is not suitable here.

      • jonsmirl
        Dec 04, 2019 @ 21:23:57

        I would not write off the analog solution so quickly. You can send a sync signal from the central SOC out to the cameras. Also, the digitizer at the central SOC is not doing any compression, it is simply digitizing the analog signal. You can also easily send the analog signal over several meters of coax.

        Or you can use the Lineni V5. It has hardware support for stereo vision and it can generate a depth map in real-time. Binocular vision works out of the box. The S3 is not designed to do this. Maybe you can force it work, but that is going to require considerable software effort.

      • jonsmirl
        Dec 04, 2019 @ 21:26:40

        Note that analog TV signal are not compressed at all.It is basically just a shift register out of the sensor chip and onto the analog line. You can get 1080P analog cameras for $10.

      • Yevgeny
        Dec 04, 2019 @ 23:12:09

        Even if you imagine that AHD can provide the quality of transmitted frames without additional noise, you still need additional equipment to capture the signal, which will increase the cost and reduce reliability. We will not get a time stamp when using AHD, and we cannot hard match paired frames. Switching an analog signal is much more expensive and inconvenient, unlike ethernet in which ten single gigabit data channels can be transferred simultaneously to one 10 gigabit ethernet port of the server.

        Thank you, I already looked at v5, this is a really powerful solution, but … my task is to centrally process various streams from a specific pool of cameras, including creating a depth map using various algorithms that can change due to progress in this scientific area. Also, the choice of algorithm can change dynamically during the execution of the main cycle, to improve accuracy or increase speed. V5 is a stereo vision task, but it does not solve all the problems of computer vision, which one such CPU still cannot handle in the pipeline, so in my case there is no sense in trying to do all the processing completely locally on V5.
        In addition, as soon as we begin to use specialized hardware processor units in our solution, we find ourselves in an undesirable dependence on the vendor, and since all processing cannot be performed on one CPU, then it makes no sense to rely on a hardware solution, in connection with the rapid development of this in this scientific field, the software solution is more flexible. I don’t think that the “BDII based on CVE” in V5 is significantly different from SGM, which is now an outsider in the ranking of stereo algorithms, while the recent development and increase in processing speed in this area is due to CNN.

        In any case, S3 is an interesting and cheapest solution and it is suitable for such simple streaming tasks.

  7. Thomas Anderson
    Dec 06, 2019 @ 16:24:34

    Will you sell any camera modules directly compatible with this? A Sony STARVIS based camera module of known quality would really help. Most things you find on the usual Chinese sites are of questionable quality (refurbished etc.). I really see this as a great way of having a reasonably more secure IP camera solution than the crap that’s usually available to buy.

    Reply

  8. tbueno
    Dec 15, 2019 @ 14:22:31

    That is great news, lots of possibilities with this chip.

    Will you make available S3 and R11 chips in your store ? Would be great to be able to source those chips from you, can’t get them otherwise in Europe :/

    Reply

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: