Low cost (Raspberry-based) particle velocimetry

This package provides a layer abstraction between both Raspberry PiCamera library and OpenPIV with the main goal of running as fast as possible in a Raspberry Pi 4. This package is meant to process the images as shown in the following image:

We have a main thread that just runs the camera capture process. This is like that due to the importance of capturing the frames over processing them in real time. Threads were selected (instead of processes) because of the shared memory to store the data between producers and consumers (easier to manage).

Once there is any frame captured (thus added to the queue -which plays the role of a pipe in our case-), the consumer starts to compute the required operations, these operations are (but not limited to):

1. Correct the distortion introduced by the lens.
2. Correct the image perspective introduced by the camera position.
3. Mask the image using computer vision technics (with morphological operations).
4. Compute the particle velocimetry using the OpenPIV library.

---

Installation

Installation expects you to use raspbian.

Before starting, we need to enable legacy mode of the Raspberry: run sudo raspi-config and then select Interface options > Legacy Camera > Enable. Then reboot to make the changes effective.

Package is not yet available in the PyPi, but it can be installed from GitHub. Firstly, we need to install some dependencies (OpenCV-related):

sudo apt update && sudo apt upgrade
sudo apt install python3-opencv libatlas-base-dev git

Once installed the dependencies, we can install the package (will install also the package python dependencies):

git clone [email protected]:valcarce01/lcpv.git
cd lcpv
pip install .

A basic test execution can be:

from lcpv.lcpv import LCPV
l = LCPV()
l.start(window_size=32, search_area_size=32, overlap=16)

Further examples can be found in the examples folder.

Run it with docker

First, install docker and add the pi user to the docker group (reboot may be needed):

curl -sSL https://get.docker.com | sh
sudo usermod -aG docker ${USER}
sudo systemctl --now enable docker

And run it!

docker build --tag lcpv .
docker run -it -rm --name LCPV lcpv \
  --resolution (1920,1080) --framerate 24 --seconds 1 \
  --camera_params src/camera_calibration_/parameters.json \
  --window_size 32 --search_area_size 32 --overlap 16

Benchmarks

Time measured (in seconds) for random images within ranges [0, 255].

Resolution	opening_filter	median_filter	correct_lens_distortion	correct_image_perspective
(540, 480)	0.01	0.03	0.01	0.003
(720, 576)	0.01	0.05	0.02	0.007
(1280, 720)	0.04	0.13	0.06	0.01
(1920, 1080)	0.09	0.29	0.13	0.02
(2048, 1536)	0.13	0.46	0.26	0.04
(2592, 1920)	0.22	0.72	0.424	0.07
(3840, 2160)	0.36	1.35	1.11	0.11