OpenMV Cam M7 first look

December 10, 2018

Initial setup guide & feature exploration for the OpenMV Cam M7. 

{ Will refer to the 'OpenMV Cam M7' as 'OMVC' from now on. }

Things you will need to start...

- OpenMV Cam M7
- microUSB to USB (normal) cable
- Male to Male pin Jumper wire
- MicroSD-Card 16gig or  32 gig

Go to OpenMV.io and download the installer.
Testing done on Windows*

Install it. I know, not ANOTHER IDE, but this one is actually useful.

The OMVC documentation and tutorials are a locally installed website. This is something I am always happy about as inevitably I will be somewhere without wifi & need detailed pinout information. Eg. in a field setting up a remote motion capture wildlife photography rig, something the OMVC is a perfect device for. Offline documentation is also crucial if you like to periodically disconnect from the internet for productivity / sanity reasons.

First boot

THE DEVICE.
Accept all the windows device driver pop-ups.

The OMVC internal memory (like Adafruit circuit python devices) will mount as an external usb drive. If you double click to open it, you can see the Main.py file on the internal board memory.

It is a good idea to read over the documentation to grasp the ways in which it operates as a usb flash drive. It's slightly different than your common expectations.

I've summarized some of the key points : 

• OMVC only works with FAT.
• Max readable SDHC card value is 32 gigs  
• If the camera saves a picture to the SD Card while mounted on a pc... the PC won't see the photo without remounting the drive as it assumes USB flash drives can't magically create files on their own (which is usually true 99% of the time).

3 ways OMVC will mount as a USB device :

1. Only internal on board memory (no SD Card) :  2 Megs
   Will appear as a standard USB drive.
   Contents will be 'Main.py' 
   
   
2. With MicroSD card inserted on board :
   To capture any images or video you have to insert a microSD card.
   You won't get a warning until the runtime file tries to save a file.
   This will now supplant the internal memory and be the default usb drive you see.
   When you plug in the OMVC, you will ***no longer see the internal 2 Meg memory*** mounted to the desktop. 
   Only the Camera SD Card will now mount to desktop. 
   
   
3. DF Firmware update
   Should only appear when two specific jumper pins are shorted.
   Very similar to updating the firmware on an Adafruit GemmaM0 or CircuitPythonExpress.
   This isn't a drive you would write anything to directly. 

Now we've got that out of the way, let's launch the OpenMV IDE !

It will likely immediately ask you to update the firmware by shorting the BOOT / Reset pins on the device with a (male-male) jumper wire.

I held the jumpers on an angle to ensure contact, if you were going to be a perfectionist, using a dual headed pogo pin adapter would be a better idea.

Follow the prompts & make sure you have a handhold you can maintain for a few minutes.

The device will need to be unplugged and plugged back in again.

When dealing with micro-circuitry, I always remove the USB cable to the computer and LEAVE the device-end inserted at all times to reduce wear and tear on the delicate board joints.

The blue LED will now flash as the firmware is updated. I didn't time it exactly, but it felt like about 30mins to re-flash the firmware. Results may vary with USB 3.0, I used a 2.0 port.

Once that is all set, you can now open the IDE and open 'helloworld_1.py' from the Example files.

Click Connect at the bottom left

The play arrow will now turn green. Click it. 

You will see the camera display on the top right of the IDE.

If not, click the arrow on the middle far right to expand the live preview & color histogram (Red Green Blue channel values).  

Let's have a look at some of the examples. 

I really have to emphasize how important reading the #comments are for each example file.

For decent results with recognition, you need good consistent lighting.

These photos are mostly screen capped from the IDE in Windows and enlarged < ENHANCE ! >, THEN compressed for the blog, so don't consider these perfect examples of the camera capture quality.

First Run Camera Tests

Shape Recognition

Circle Detection

Hair Elastic

Rect(angle) Detection

Real time Edge Detection

Thankfully, It did not recognize it's own reflection, or demonstrate any other signs of self awareness, so we are safe for now.

HOWEVER it does have the capability to learn to recognize things, utilizing python machine learning, so that might be a fun future experiment.

 

Facial Recognition 

Facial detection works surprisingly well in properly lit conditions.

The default example file was able to identify two faces with different skin colors moving slowly in the same frame at the same time. You do need to have both eyes facing the camera, but there were surprisingly low false positives.

Facial recognition doesn't do as well in poor lighting. One could tweak the code tolerance values, but for low light uses, the optional Infra-Red lens would make more sense. This blog is more of a cursory run through the example files.

Conclusion 

Barely scratched the surface of this exciting new platform and already very impressed. This is the best option if you want to learn camera based programming.

According to the documentation, the IDE previews camera frame rates at about half the capture speed. SDCard recording is at full frame-rate. Perceptually, I didn't notice any frame-rate loss on my laptop.

If you want to look at the python example files directly, the Examples are buried here in windows (will very based on OS):
C:\Program Files\OpenMV IDE\share\qtcreator\examples

One more bonus to using micro-python vs. C is the next time you plug in your M7 board (without the sd card inserted), you can open & read the .py file to remember what program the camera is running.

This is a huge improvement over the more traditional Arduino approach where you can't easily read the sketch that was previously written to the board.

 The M7 captures at 640x480 but depending on the use-case scenarios sometimes what looked like 320x240 greyscale to maintain real time recognition performance for barcode or facial recognition. I'll have to dig into the documentation and do further testing before I can confirm that.

The only things close to this at the moment would be either 

• A RaspberryPi 3b+ with  Camera and a lot of python machine vision libraries installed (you would have to set up on your own).
  
  
• Or the Google AIY Vision Kit which relies on the google cloud to do the heavy lifting. This ties a raspberry Pi into the google cloud & needs always on wifi to function. Additionally, TensorFlowSDK is a relatively extensive API. Still a cool kit though & can be found here on the elmwood store.
  
  
• Arduino does have serial cameras, but wouldn't have the processing to do any recognition type tasks. You would need a co-processor board and at that point you should look at OpenMV anyway. 

In further blogs, I will be exploring the various shields the OMVC can interact with (LCD, wifi, etc.).

To further explain one point, micro-python is a smaller, reduced function subset of the desktop and server language Python. Micro-Python is made specifically for single board circuitry devices.

The OMVC device works with Micro-Python, not normal Python, so installing desktop python should be redundant as OpenMV IDE should install all you need.

However, if you have NO standard Python on your system already, and at some point later on you want to learn more standard Python to improve your micro-python skill-set, it is generally recommended to go with the latest version (3.4 as of this blog).

---

Also in News