Raspberry Pi control of photogrammetry rig
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.idea
cameractrl
cloud_drive
configuration
docs
restapi
rpihat
tests
venv
website
.gitignore
README.rst
requirements.txt
rig_runner.py
update.sh
util.py

README.rst

[bp100a] Raspberry Pi Photogrammetry

Overview

Code for controlling a photogrammetry device using a Raspberry Pi. The physical rig is 3D printed from this Thingiverse model Photogrammetry.

Just a note. This is a work in progress. What you see is a moving target both in terms of software and hardware. The 3D printed rig is unstable, I'm working on ways to improve it's rigidity. The software is an MVP, it works but there're no unit tests and it needs major refactoring. But it works, which is way better than a beautiful design that doesn't.

How it works

Written entirely in Python 3.6 and hosted by a Raspberry Pi 3B running Stretch-lite (no GUI).

rig_runner.py: This is the main task that runs the rig and is responsible for driving the stepper motors and taking the pictures. Actions are communicated to/from using a set of Beanstalk tubes (queues). This process is responsible for homing the rig (stepping to end stops), calculating the stepping points for photos, as well as interfacing and controlling the camera.

Based on the RaspiMotorHAT shield, borrowing heavily from the python sources.

Pretty basic structure, there's a "while" loop that polls the beanstalk tube and then processes the task. When stepping the motor it's a much tighter loop, but there is a "yield" function to check for exit conditions (like end stop switches).

rig_control.py: This is a Flask REST API that sends tasks to the rig_runner process via beantstalk tubes. The REST API is called from a set of HTML pages that comprise the UI for the site. The Flask WSGI server is hosted by Gunicorn with NGinx proxying all http requests. So not much going on here.

Once again beanstalk is a pleasure to work with, much simpler and easier to implement than other queues. For this implementation I was hoping to keep setup & deployment as simple as possible, beanstalk did it's best in supporting that goal.

google_drive.py: This is a separate process that uses a beanstalk tube to receive authorization tokens and photos to post to the Google Drive.

You must specify a google drive, the application will create a /rpipg folder in the root of your google drive. Under that, for each modeling session a folder with the date/time when the modeling session starts is created for all the photos.

beanstalk: This is a simple, effective queue. On startup I set the job size limit to 10MB, which is atrocious but given this is a dedicated device & app forgivable. Since the photos are about 4.5MB (base64 encoded), if the queue backs up there'll be issues.

website: The website is based on a template from the Envato market called Kolor. The crown jewel of the website is the circular slider I found called roundSlider. It's totally awesome and the structure of the Kolor templates makes it easy to integrate.

https://raw.githubusercontent.com/bp100a/RpiPG/master/docs/scan_screenshot.png

I'm using jquery/ajax calls to read/write the REST APIs, for now you can "home" and initiate a "scan". There is a "status" API that I poll that basically displays internal status of the rig controlling software to a jquery Ticker (I need a new ticker, this one doesn't work quite how I want it to).

I'm using Nginx to host the website, which I find far easier to work with than Apache (as in easier to understand configuration).

camera: Finding a camera that would work with the gphoto2 software library (and wasn't a $600 DSLR) was a challenge. I'm using the GPhoto2/LibGPhoto2 library interfacing a Nikon Coolpix S3300. I started with v2.5.17 but turns out there's a "read manifest" call that isn't supported by the S3300 so every picture encountered about 50 secs of timeouts. Upgrading to v2.5.20 fixed that, now pictures take about 16 secs each, which is tolerable.

I tested about 5 cameras (several I had on hand, and two I purchased on eBay) before I found one that worked. The S3300 is compact and has good resolution, and best of all can be powered via the USB cable. The downside of the S3300 is that while in PTP mode you can't use the viewfinder or any camera controls to "frame" the image being photographed.

https://raw.githubusercontent.com/bp100a/RpiPG/master/docs/S3300.jpg

Regards,

Harry