project-image

OpenMV Cam H7 - Machine Vision w/ MicroPython.

Created by OpenMV

The OpenMV Cam H7 is an open-source MicroPython powered machine vision camera designed for low-power real-time applications.

Latest Updates from Our Project:

Backer Kit Survey has gone out!
over 5 years ago – Sun, Nov 11, 2018 at 09:49:52 PM

Hi folks,

Just a reminder that the Backer Kit Survey has gone out. We've received about 75% of all backers completing the survey. Please complete your survey as quickly as possible. Backer Kit will keep emailing you until you finish it.

We have a good idea of the production volume of everything right now but would like to have this over 90% before we place medium-lead-time part orders.

In the mean-time, we've sent out our order for pre-production prototypes to test out the final production process and hope to get these back in a month or so. After confirming our pre-production prototypes we'll get all the PCBs and short-lead-time parts ordered.

Anyway, more feature update posts coming soon once we get through all manufacturing details!

Backer Kit Survey going out soon!
over 5 years ago – Sat, Nov 03, 2018 at 06:56:21 PM

Hi folks,

We've finished setting up our Backer Kit pre-order store and Backer Kit survey. The pre-order store is live now and the survey will be going out shortly to everyone. Backer Kit requires us to send out a "Smoke Test" first to just a few backers so it won't be till next week that everyone receives the survey.

Anyway, in other news, Kickstarter has collected the funds raised and we'll be able to send our PO to our contract manufacturer Etonnet soon. There's still some work to finalize all the files for hand-off to Etonnet but we hope to complete this process by next week.

Also, we got an update from Etonnet about the STM32H7 chip delivery schedule. Arrow who is fulfilling the 10K order has notified us that the chips will ship to Etonnet in early February 2019. Etonnet will buy all other parts and PCBs such that they can start assembly of the cameras in March after the Chinese New Year. We already have all the testing and programming jigs in place from our last OpenMV Cam M7 model so production should only take about two weeks in March. Things look good for us to start shipping in early April. If Arrow is able to deliver the STM32H7 chips earlier we'll be able to ship sooner. The STM32H7 chip is the only long-lead time part.

In the mean-time we'll continue selling he OpenMV Cam M7 and maintain stock of it until next year where we'll switch production over to the OpenMV Cam H7.

BackerKit Setup
over 5 years ago – Fri, Oct 19, 2018 at 08:20:27 PM

Hi folks,

First, thanks for helping us raise $95K to fund the OpenMV Cam H7! Awesome! Second, I was at ARM Techcon this weekend and ran into Massimo Banzi, the founder of Arduino! He likes the product too!

Anyway, I'll be setting up Backer Kit this weekend to get the pre-order store operational along with dealing with a few other things and then sending out a survey for you to fill in what you actually ordered.

24 Hours to go!
over 5 years ago – Tue, Oct 16, 2018 at 08:56:49 PM

Hey everyone, with 24 hours left in the campaign, we would like to thank you for all your contributions to the OpenMV project. We have raised about $88K from almost 600 backers! Thanks to you all, we will continue to explore and develop new machine vision solutions for beginners and experts alike. Note if you would like to share the project one last time, or adjust your pledge, now would be the time to do so.

In other news, OpenMV is going to be on this year's Hackaday Prize judging panel. It's worth mentioning that the OpenMV project began as a Hackaday Prize entry! Best of luck to all participants and congrats to Kwabena Agyeman for being selected as a Hackaday Prize judge.

More CNNs on the OpenMV Cam H7
over 5 years ago – Mon, Oct 15, 2018 at 12:18:28 AM

Hi folks,

First, thanks so much for backing our Kickstarter! Please continue to tell your friends and family about our project's last few hours fundraising on Kickstarter!

New CNNs

I spent this weekend working on a bunch of new nets for the OpenMV Cam H7. The training formula on how to go about making your own nets is pretty solid. Once you have a deep learning rig with an NVIDIA gpu it's quite straight forwards with our guide. However, training nets can be a challenge. It's easy to get a high accuracy on your training and test datasets but fail to have any accuracy in the real-world due to various issues related to how your net was trained. So, while not an expansive as I hoped, I got one nice network working on the OpenMV Cam H7 over the weekend which solves letter recognition for folks:

The Chars74K CNN can be targeted at any ROI on the image [x, y, w, h] so you can use it to read text with your OpenMV Cam H7 that's in a fixed position away from the camera. Let's say you need to read a mechanical meter with the OpenMV Cam H7 - with the Chars74K network this is easy now. You can target the network on fixed position letters/numbers and build up a string of what is in front of the camera.

Moving on, we'll have a few more out-of-the box CNNs for folks to use with their OpenMV Cam H7. That said, if you're serious about CNN support you have to actually capture source images/video for the particular application your interested in with your OpenMV Cam (or IDE) and train a net on those images for the best results. Things like image scaling issues and class imbalances can make a network poor for the application at hand if you're trying to target one dataset for another application it wasn't created for. Good dataset preparation makes the difference between getting a working CNN at the end of the day and a broken one.

Anyway, Thanks for reading!