Hi. I'm greatly interested in DepthAI USB3 FFC Edition. And I'm also interested in both the color camera and the shutter pair cameras. I intend to use these for an open source, DIY 3D printed pair of eyeglasses that I've been developing for the blind. I've been completely blind myself for the past 17 years. Here's a video of my preliminary prototype in action, though it's qquite overtly bulky at the moment: https://www.youtube.com/watch?v=nVkaMl6um9A Anyway, I'd greatly appreciate some help regarding my questions here: 1. Would it be possible for the lens of the color camera and the shutter pair cameras to be separate from its board? • Currently, these MIPI CSI-2 cameras have the same dimensions as RPi Camera v2.1 (25x25x9mm) ; • This means the lens of these cameras are attached to a 25x25mm board ; and • It'd be really helpful if there are versions of these cameras where the lens is separate from the 25x25mm board via a flex PBC, something similar to this USB camera here, though with a 2x or so longer flex PBC >> http://www.camera-module.com/product/8mpcameramodule/8mp-usb-camera-module-sony-imx179-auto-fixed-focus.html 2. As a follow-up to my first question — Are there other MIPI CSI-2 cameras (color camera and shutter pair cameras) that are designed this way, which would most likely work with DepthAI USB3 FFC Edition? • I can't seem to find any. 3. And lastly, what are the dimensions (length x width x thickness) of the DepthAI USB3 FFC Edition? • I can't seem to find this anywhere. I'd greatly appreciate some assistance. Thanks in advance! :)

So this thread ended up in email as well (where we've gone back/forth some). Cross-posting from email to here: > Hi. We're building a pair of 3D printed eyeglasses for the blind (I'm > completely blind myself). > And we're interested in the DepthAI USB3 FFC Edition. > Plus, we want to use it along with both your 12MP color camera and the > global-shutter stereo pair. > But we have the following questions: > 1. What are the dimensions (length x width x thickness) of the entire board? > 2. All cameras (color camera and shutter pair) each has a similar dimension to > the RPi Camera V2.1, correct? (25 x 25 x 9 mm) > 3. So this means the lens of each camera is attached on top of a 25x25x9mm > board, correct? > 4. If yes, then would it be possible to have the lens of each MIPI CSI-2 > camera separate from its board through flex PBC? Something similar to this USB > camera, though with a 2x or so longer flex PBC? >> > http://www.camera-module.com/product/8mpcameramodule/8mp-usb-camera-module-sony-imx179-auto-fixed-focus.html > Our purpose in asking is that we want to integrate the cameras into the front > frame of a 3D printed pair of eyeglasses. > We want to position the lens of the left and right cameras of the shutter pair > either at the ends of the front frame, or at the start of the left and right > arms of the glasses. > Meanwhile, we want to position the lens of the color camera at the center of > the front frame (the area of the user's nose bridge). > Please advise. Thanks! :) 1. So the BW1098FFC kit is composed of 4 boards total to allow modular installation/prototyping like this: a. The Myriad X central board. This is the BW1098FFC itself. It is 60mm x 40mm, with mounting holes spaced at 34mm x 54mm. The hardware design for this is [here](https://github.com/luxonis/depthai-hardware/tree/master/BW1098FFC_DepthAI_USB3). b. The Color camera board. This is the BG0249. It is 25mm x 21.86mm. The board dimensions, hole pattern and sizes, and center of the image sensor are identical to the Raspberry Pi v2 dimensions, here. We used a different FFC though as the camera/control is very different. The hardware design for this is [here](https://www.raspberrypi.org/documentation/hardware/camera/mechanical/rpi_MECH_Camera2_2p1.pdf). c. The grayscale camera boards. Two of them for stereo depth. These are the BG0250TG. Also 25mm x 21.86mm. The board dimensions, hole pattern and sizes, and center of the image sensor are identical to the Raspberry Pi v2 dimensions, here. We used a different FFC though as the camera/control is very different. The hardware design for this is [here](https://github.com/luxonis/depthai-hardware/tree/master/BG0250TG_DepthAI_Mono_Camera). 2. Yes, that's right. The dimensions of the Raspberry Pi camera are [here](https://www.raspberrypi.org/documentation/hardware/camera/mechanical/rpi_MECH_Camera2_2p1.pdf). Our thickness is different though because we use different camera modules. The 3D STEP files for our cameras are available in the links above. And all our hardware is [here](https://github.com/luxonis/depthai-hardware). 3. The Raspberry Pi dimension document ([here](https://www.raspberrypi.org/documentation/hardware/camera/mechanical/rpi_MECH_Camera2_2p1.pdf)) specifies where the center of the images sensor is located. So from that, the center of the lens is 9.462mm from the edge of the board with the FFC connector and 12.5mm from the other side. 4. This may be possible, but we are very constrained based on supply chain permutations. So right now we are using a camera module on these modular boards that has a ~0.75 inch long FFC. There may be others with longer FFC, but we are unaware of them. The easiest thing for a prototype would be to use our modular boards above with our FFC. It's wider than what you show by probably 2.5x (so about 1/2" wide). But for a prototype should be doable. The frames of the glasses could just be thick to accommodate. Since we are open source, one could take our design and change it to use a narrower FFC as well. And if that's of interested, we are happy to help facilitate. There is the blind institute of technology close to us in Denver. We would be happy to work with you and them to help make this happen. Still it would be better to prototype initially with what we have now, even though the glasses would be bigger than they could be after such optimizations. So what you mention you want to do is doable now with the BW1098FFC kit. As the cameras are modular and come with a 6" flexible flat cable each. So it might look ugly to start... but it should make a viable proof of concept. And we can help make custom versions of these boards to make this fit better in glasses. What would likely be best is a single PCB that holds all 3 cameras on the front of the glasses. So that way not cables to wrangle. Thoughts? Thanks, Brandon

I like the back-of-neck idea for processing to start with. In terms of keeping costs down, I would say the B1098FFC kit itself is what should be used. https://shop.luxonis.com/products/depthai-usb3-edition Longer FFC cables can be used up to 12 inches... just let us know if you plan on ordering that and we'll order the extra-long FFC (12" instead of 6" cables) and included them in the box. With those I think nape of neck would be supported. I'm going to introduce you to blind institute of technology now. Maybe they can help with funding. > Hi Brandon. Good morning! :) > > This is just simply amazing — Thanks for supporting the project! :) > • Hopefully, we'd be able to get additional support for the custom > PCB from the Blind Institute of Technology and others ... > > Yes, DepthAI USB3 Onboard Camera Edition sounds promising! :) > • 7.5cm spacing for the grayscale cameras, and the color camera in > the middle, look just about right ; and > • My blind friends and I also discovered through field tests that > cameras equipped into eyeglasses need to be tilted downwards at around > 12 to 15° ... > — This tends to correctly simulate what people with eyesight see (and > what they naturally focus on) in front of them ; and > — Not doing this though tends to capture a lot of things above the > normal field of view ... > > Oh also, are the cameras of BW1098OBC embedded unto the board? > Or does it have long FFC connectors running to the board's MIPI CSI-2 ports? > • I'm just thinking if it does have long FFC connectors, then the > board itself (and the heatsink) can hang somewhere at the user's nape. > And this in turn can be connected via a long USB cable to a > battery-powered host, which can be say clipped unto the user's belt / > waist, or inside the user's pocket ... > > Eitherway, I'm really energized to be able to significantly enhance > the product with DepthAI! :) > • Right now, I'm using a single color camera and an ultrasonic sonar > ranger ; and > • Unfortunately, this only detects distances of objects right in > front of the user, as the sonar sensor sits below the color camera in > the middle of the glasses ... > — This also forced me to do a workaround in OpenCV where 2D X and Y > coordinates of detected objects are used to tell the user where the > object is in 6 quadrants (3 upper quadrants [left, center and right], > and 3 lower quadrants [left, center and right]) ; and > — Another thing I found interesting though is that blind users (well, > me and almost all of my blind peers in the communities I've been > participating in) prefer clockface directions when being told the > location of objects, i.e. 10 o'clock for objects in upper left > quadrant, 2 o'clock for upper right quadrant, etc. (we're also > actually trained this way during mobility rehab after I lost my > eyesight) ... > > I really do hope we can continue working together on this. :) > Thanks again! This is really awesome! :) Very good to know on the 12 to 15 degree down-tilt! The cameras on the BW1098OBC are embedded directly on it. The OBC actually stands for 'On Board Cameras'. And that's good to know as well for clockface directions. That makes a ton of sense! I think in tactical situations that is the preferred terminology as well. It's fast/simple and easy to understand. Thanks again!

Hi again Brandon! As a working proof of concept for this, I was just thinking about the complete DepthAI edition with the RPi Compute Module 3+ and 3 cameras ... • Are the dimensions of these the same as the RPi Compute Module 3+ (671x310mm)? I can perhaps do the following things: 1. Connect a LiPoly battery and 3D print a case with something to attach it to the front neckline of a shirt ; 2. Connect Bluetooth earbuds ; and 3. Code something up to deliver audible info to the blind. Thoughts? Thanks for your time! Greatly appreciate it. :)

Hi @"MarxMelencio"#30 , Very happy to help. Really want to make this happen! I pinged some other to see if they'd be interested in helping on this as well. So the dimensions of the Raspberry Pi Compute Module Edition of DepthAI (Part Number: BW1097) are 106mm x 60mm (and ~37mm tall, from the tallest component on the bottom to the tallest on the top, both of which are heatsinks). So it holds on the back of it a RPi Compute Module 3+ which has the dimensions you mention. 1. Yes. This is a great idea as then the whole computer is integrated. 2. So this model does come with WiFi but not bluetooth. I'm going to look if there is a USB bluetooth that you cold plug in to give bluetooth audio. Any one that works with the Raspberry Pi should work. 3. Yes! Would be so awesome. Overall I think this is a great idea for getting off the ground. So actually a GoPro chest-strap would be awesome for holding it. That way it doesn't bob up/down when walking. https://www.amazon.com/gp/product/B01171X0UW/ref=ppx_yo_dt_b_asin_title_o08_s00?ie=UTF8&psc=1 We bought this and although we haven't tested the chest strap, it looks great - and the whole kit is only $25.99 with good reviews. So Martin Nossel made a case for the DepthAI Compute Module Edition that could be used and make a GoPro mount for the back of it. Right now it is designed to mount to a seat post. https://docs.google.com/document/d/1Ds81P4aJZK22VdSS5-GB5c5KEfIcZAO6Q5mYQqgXFFc/edit I just asked a mechanical engineer friend if he would be able to help make this. Thoughts? Thanks, Brandon

@"Brandon"#4, Sounds really cool! :) 1. Yes. Also, I don't think the device bobbing up and down while walking would be a show stopper. Though of course, it would be fantastic to at least significantly minimize this from happening ... • That's because when my blind friends and I tested my current implementation, we found out that we mostly need options to be on-demand ; • So I used a single Bluetooth button and coded something that can allow the user to select an option on demand, i.e. Detect objects / persons, detect central distance (using the sonar sensor), read text, go down each line of recognized text, request for manual visual assistance through Skype / Facebook, play multimedia like an audio book or so, etc. ; and • So this time, I'm hinking of connecting to the device itself a tiny programmable USB or GPIO button, in order for the user to select / deactivate an on-demand option ; and • But a GoPro chest strap, or something at the back of the DepthAI Compute Module Edition that can securely fasten it to the neckline of a user's shirt, would be awesome ... 2. Yes, connecting a tiny USB Bluetooth dongle is a good solution. 3. Yes. Audible output is quite straightforward through the Pico TTS library in Python (there are other libraries, though this is the one that my blind peers and I really like) ... Overall, a case to hold the shutter camera pair (left and right), the center color camera, the DepthAI board, the RPi Compute Module 3+, a rechargeable LiPoly battery, the USB Bluetooth dongle, a tiny programmable USB / GPIO button / switch, and clips at the back to attach it to the neckline of a user's shirt would be really cool! :)

All sounds great. I think this is a good plan and agreed with what you propose for the prototype. Will see if someone would be interested in making this case. On the Bluetooth adapter for Pi, this one on Amazon says it works with the Pi, [here](https://www.amazon.com/TECHKEY-Bluetooth-Headphones-Raspberry-Compatible/dp/B07J5WFPXX/ref=sr_1_1_sspa?dchild=1&keywords=bluetooth+dongle+raspberry+pi&qid=1590543891&sr=8-1-spons&psc=1&smid=A3KHUM65I50XQR&spLa=ZW5jcnlwdGVkUXVhbGlmaWVyPUEzQ0RCNzdHSktUNjVIJmVuY3J5cHRlZElkPUEwMDQ5Mzk0M002N0pBSFBZU0tNWiZlbmNyeXB0ZWRBZElkPUEwNjU3MDQ3MTQzNk80Mk8yWTI0NSZ3aWRnZXROYW1lPXNwX2F0ZiZhY3Rpb249Y2xpY2tSZWRpcmVjdCZkb05vdExvZ0NsaWNrPXRydWU=) Also have you read about Unfolding Space by Jakob Kilian? It uses a depth sensor and a glove to provide feedback: https://www.designboom.com/technology/jakob-kilian-unfolding-space-see-with-your-hands-09-14-2019/

Hi @"Brandon"#4, Yeah, that thing in Amazon looks good. :) • Yes. I think that's small enough. But still efficient ; and • I have a [URL="https://www.amazon.com/Logitech-K375s-Keyboard-Connectivity-Bluetooth/dp/B07387XSXR"]Logitech K375 Bluetooth keyboard[/URL]. And it came with a USB dongle. I think just about the same size (can't find it now, though I remember) ... And yes. Though I haven't used Unfolding Space yet. But I think it's really cool! :) • I really think haptic feedback in this use case should be something explored more ; and • I've experimented a little about this, i.e. Coordinating [URL="https://www.seeingwithsound.com/meijer.htm"]Peter Meijer'S[/URL] vOICe (also known as [URL="https://www.seeingwithsound.com/"]SeeingWithSound[/URL]) through distance data of central objects from a [URL="https://www.robotshop.com/en/devantech-srf10-ultrasonic-range-finder.html"]Devantech SRF10 ultrasonic ranger[/URL] and an [URL="https://www.adafruit.com/product/1201"]ERM vibrating disk[/URL] connected to a GPIO microcontroller called [URL="https://www.ti.com/product/DRV2605L"]DRV2605L[/URL] from Texas Instruments, running in an RPi Zero, and also in an X86 SBC called [URL="https://www.ecs.com.tw/ECSWebSite/Product/Product_Overview/EN/System/LIVA%20Q/LIVA"]Liva Q[/URL] ...

Thanks. All looks good to me. :-) One thing we haven't made run yet is OCR. I think it's doable we just haven't done it yet. The default OpenVINI model is humungous for some reason, so will need some looking/iterating.

Hi @"MarxMelencio"#30 I'd love to help your prototyping. I am a principal mechanical engineer at Amazon. I previously prototyped a wrist wearable for low vision people with movidius onboard (home project). https://bit.ly/2XEl0A8 Andre

Thanks @"andredavidbrown"#p185 ! Really like your work there. Super fun! Definitely we're excited to help out on all of this. So the System on Module ([here](https://shop.luxonis.com/products/bw1099)) allows integrating the depth and AI capability into a custom formfactor easily. And all the baseboards for it are opensource so are a great starting point for doing a custom formfactor: https://github.com/luxonis/depthai-hardware

Questions Re: DepthAI USB3 FFC Edition & Cameras

MarxMelencio

Hi again @Brandon,

DeepWay looks like a really cool project — Thanks for the heads up! 🙂

This gives me some ideas to, in future versions, integrate vibrating ERM motor disks into the backside of the ID enclosure concept, so as to provide real time haptic feedback. 🙂
• Users will feel on their chests where detected object's are in any of the 6 quadrants I previously mentioned (top left, top center, top right, bottom left, bottom center, bottom right); and
• That's aside from providing us with open ear audible input for the class, clockface location and distance of detected objects, along with other beneficial functions like OCR and so on! 🙂

Brandon

MarxMelencio Very Cool!

Neat to see you helping in so many ways!

Also, would it be possible for me to order a BW1098OBC that's already switched to use USB3-C power and has the heatsink taken out?

Yes, we can definitely do this. Also one of my friends who's a mechanical engineer may be able to help to make a heat-spreading case for this design.

If you order the BW1098OBC (here), we'll ship you one with the power switched to USB input and also have the heatsink removed (but included in the shipment).

Be sure to not run the device long with the heatsink removed though without another equivalent heatsink (e.g. the heat-spreading case) in its place, as it will very likely overheat.

And WRT customs: let us know if there's anything we need to do to help. Usually these need to be cleared by the receiving party though.

Thanks again,
Brandon

MarxMelencio

Hi @Brandon,

Thanks! 🙂
I'll order from that page for sure, as soon as I raise the money for the unit and shipping, hopefully later this week.

Yes, I'll let you know as soon as DHL gets back to me.
Their tracking page says the package is already here in Manila, so I contacted their Manila support directly, since it's stuck at "customs status updated".
But they haven't gotten back to me yet - I'll call them again tomorrow morning GMT+8.

Thanks again, and have a great week ahead! 🙂

Brandon

Thanks @MarxMelencio !

So for anyone else following the thread, we have moved a lot of the iteration to our slack community channel on this ( #visual-assistance ), here.

And we have two parallel efforts in progress:

Prototype with existing DepthAI hardware (BW1097 and BW1098OBC in process, as above).
Build a custom completely-integrated system for this visual assistance device, specs of which are here.

Please feel free to come join us in slack (and contribute) if this is of interest to you!

Thanks,
Brandon

Brandon

So I made a Github issue for the hardware design to make it easier to discuss/iterated on/etc. (than the Google doc):
https://github.com/luxonis/depthai-hardware/issues/9

Brandon

So @MarxMelencio - Rares at Luxonis was able to get the text detection running on DepthAI even though OpenVINO says that it doesn't work. So this is a great step for then being able to do OCR on the text.

Github issue here

MarxMelencio

Hi @Brandon,

Cool! 🙂
• I'll take a closer look much later this week. Planning to test some sequential methods for running object detection, tracking, text detection, and text recognition. Will test OpenVINO, Pytesseract, and Microsoft OCR Read Cloud API ...

Also, I managed to successfully run BW1097 with my PiJuice HAT and 40-pin GPIO extender ribbon cable (male to female).
• However after a couple of days, PiJuice suddenly isn't able to supply enough power. Even if the PiJuice HAT is still working, based on its LEDs.
• And my Zero2Go Omini isn't working. I remember we used a capacitor to make this work with an RPi Zero WH ...
• So I'm back to using BW1097's official power supply, while waiting for the USB-A male to barrel plug that I ordered, as we discussed in Slack ...

And I'll in the next day or so share a video in Slack regarding some things I've been doing.
• This mainly involves calculating object placement in clockface locations, i.e. If x_distance and y_distance <= -0.85m, then clockface location is "7 o'clock" (Python treats negative values with higher numbers as bigger, instead of smaller, so this is >= in my code) ...
• Converting z_distance into number of standard steps, i.e. If <= 1m, then "2 steps away or less" ...
• Stringing this up with mobilenet object class, i.e. 9 o'clock. Person. 3 steps away ...
• Generating quick beeps with stereo panning to prefix TTS of combined result, i.e. Quick stereo pan from right to left if object is at the left (10, 9 and 7 o'clock), no stereo panning for center clockface locations (12 noon, dead center, 6 o'clock), etc.
• Also using fade ins / outs and 3 different pitches for these beeps, i.e. Lower pitch for 5 / 6 / 7 o'clock positions, and so on, to optimize much faster user recognition of object location ...
• Plus, using 3 different TTS voices for the same reason, i.e. One voice for object location, another for object class, and another for distance ...

Cheers! 🙂

Brandon

Also, I managed to successfully run BW1097 with my PiJuice HAT and 40-pin GPIO extender ribbon cable (male to female).
• However after a couple of days, PiJuice suddenly isn't able to supply enough power. Even if the PiJuice HAT is still working, based on its LEDs.
• And my Zero2Go Omini isn't working. I remember we used a capacitor to make this work with an RPi Zero WH ...
• So I'm back to using BW1097's official power supply, while waiting for the USB-A male to barrel plug that I ordered, as we discussed in Slack ...

Interesting. Checking out the PiJuice HAT specs now. FWIW I did order the USB-to-barrel-jack cable here and tested it, and it does work. It also looks identical to the standard power supply barrel jack, so that's good.

Output amperage - maximum current at 5V gpio is 2.5A and at VSYS output 2.1A, but also this depends heavily on battery capacity. For BP7X have measured around 1.1A at 5V GPIO and around 1.6A at VSYS output. Obviously, this also depends heavily on the current draw demanded by the Raspberry Pi/device itself. To achieve a maximum of 2.5A it will need battery over 3500mAh.

What size battery are you using with the PiJuice? From their specs here it seems 3.5Ah or higher is required.

And I looked at the Zero2Go Omni... from the look of the board I don't think it can provide the current. Checking to see what its specs are in terms of output current. It says Output 5V (±5%), maximum 2A but it just doesn't seem like it has enough onboard capacitance/energy storage to reliably support that for a dynamic load (like a computer system). So this fits w/ what you were saying about having to add a capacitor. It looks like it's more intended for <1A use.

• I'll take a closer look much later this week. Planning to test some sequential methods for running object detection, tracking, text detection, and text recognition. Will test OpenVINO, Pytesseract, and Microsoft OCR Read Cloud API ...

Super excited to learn what you find! We have not researched this field much. The closest I've done is OpenALPR on an iPhone. :-)

And really cool on the effort you've done to share information in clock direction and w/ audio feedback w/ different pitches/etc.! That's super neat! Really excited to hear it in action!

Thanks again,
Brandon

Brandon

We have the first version of the visual assistance variant working now! Some photos below show that it is small enough to be wrist-worn, measuring 60mm x 47mm x 27mm.

« Previous Page