1 week ago

Monitor plant growth with AI and OpenCV

Measure the height and width of objects with your Raspberry Pi using only the Pi Camera Module and OpenCV

This is a static object that we have measured. By comparing with the plant’s size, our code can estimate the height 

Do you know how tall your plant is? Do you wonder how high it is growing? Does using a ruler just sound like effort? Well, we have the tutorial for you. We’re going to monitor plant growth using images taken with a Pi Camera Module. In the process, we’ll introduce you to OpenCV, a powerful tool for image analysis and object recognition. By comparing your plant to a static object, OpenCV can be used to estimate its current height, all without touching. By adding data from other sensors, such as temperature or humidity, you too can build a smart plant.

This tutorial was written by PJ Evans and first appeared in The MagPi 81. Sign up to our newsletter to get a free digital copy of The MagPi every month. Or subscribe to us in print (12-month subscribers get a free Raspberry Pi computer).

Pick a spot for your Plant AI monitor

We’re going to be using OpenCV (version 2) to measure our plant. To make sure our measurement is as accurate as possible, several factors need to be considered. The background must be plain and contrast against the plant; black or white is ideal. The area you are using must be very well lit when the image is taken. Finally, the camera must be 90 degrees to the plant, facing it directly. Any angles on any axis will cause poor measurement. If you are mounting the camera at a significant distance, you may want to consider a telephoto lens.

AI Plant Monitor at 90 degrees

Our testing rig. Note the camera is as close to 90 degrees to the plant as we could manage

Get a plant and object

So how do we pull off this measuring trick? We can estimate the height of an object by comparing it with another object that is a fixed and known width and height. The image taken by the camera needs to include both objects. Get a rectangular object of a comparable size to your plant; we used a long piece of Lego. A metal ruler would also work well. Mount your object in front of the camera so it is on the left of the image and the same distance from the camera as your plant.

Prepare your Pi

The software used to analyse the image is the powerful OpenCV library and its Python bindings. You can use any current Pi for this project, but the higher-end 3B/3B+ will be much quicker at processing the image. OpenCV requires an X Window system in place, so we need to start with Raspbian Stretch including the Raspberry Pi Desktop. Once ready to go and on the network, make sure everything is up-to-date by running:

sudo apt update && sudo apt -y upgrade

Finally, install the camera, if you haven’t already, and enable it in Preferences > Raspberry Pi Configuration > Interfaces, then reboot.

Install dependencies

OpenCV is a bit trickier to install than most packages. It has lots of dependencies (additional software it ‘depends’ on) that are not installed alongside it. We also need a few other things to get OpenCV talking to the camera, so open up a Terminal window and run the following commands:

sudo apt install python3-pip libatlas-base-dev libhdf5-100 libjasper1 libqtcore4 libqt4-test libqtgui4

pip3 install imutils opencv-contrib-python picamera[array]

Although piwheels speeds things up considerably, you can expect the second command to take a little while to run, so now’s a good time to get a cup of tea.

Calibration test

Before we start running code, make sure everything is lined up as you would like it. The easiest way to do this is to open a Terminal prompt on the Desktop and run the following command:

raspistill -t 0

This puts the camera in ‘preview’ mode (you’ll need to be directly connected to see this – it won’t work over VNC without extra configuration). Study the image carefully. Is the camera at an angle? Is it too high or low? Is there enough contrast between your object, plant, and background? How’s the light? Press CTRL+C when you’re finished.

Coding time

Enter the code from the plantbot.py listing in your editor of choice and save it. If you don’t fancy typing, you can download the code and some sample images from here. The code takes an image from the Pi Camera as a stream and sends it to OpenCV. The image is then placed through some cleaning filters and the shapes detected. We know the nearest shape to the 0,0 ‘origin’ (the top-left of the image) will be our calibration object. So long as we know the width of that object, we can estimate the size of others.

OpenCV calibration test

An example of what OpenCV sees, with the calibration object on the left, and some daffodils on the right

Run the code

Let’s take our first image. Carefully measure the width of your calibration object. Ours was 16 mm. To run the code, enter the following command in the same location as your plantbot.py file:

python3 plantbot.py -w 16

Replace the ‘16’ with the width of your calibration object. After a few seconds, you’ll see the image with, hopefully, a light blue rectangle around the leftmost object and one or more boxes around the plant. As the plant is an irregular object, OpenCV will find lots of ‘bounding boxes.’ The code takes the highest and lowest to calculate the height.

OpenCV mapping the test image

This overlay shows how OpenCV has mapped the test image. It can be far more accurate, but this faster method meets our needs

Big plant data

Once you’ve had fun measuring all the things, you can alter the code to write the results to a file. Add the following to the ‘import’ section at the start of the code:

import datetime

Now remove the final six lines (the first one starts ‘print’) and replace with:

print("\"" + str(datetime.datetime.now()) + "\"," + str(plantHeight))

This version can be employed to create a CSV file of measurements that may be used to create graphs so you can track your plant’s growth.

Keep on schedule

Let’s say you wanted to take a measurement every day. You can do this by using Raspbian’s cron system. Cron is an easy way to run scripts on a regular basis. To access the main cron file (known as a crontab), run the following:

sudo nano /etc/crontab

At the bottom of the file, add this line:

0 14 * * * pi python3 /home/pi/plantbot/plantbot.py -w 16 >> /home/pi/plantbot.csv

This tells cron to run our script every day at 2pm (that’s the ‘0 14’ bit) and append its output to plantbot.csv, which can be read by any popular spreadsheet app.

Got water?

Now you’re collecting data on your beloved plant’s height, why not add other data too? A common Raspberry Pi project is to use soil moisture sensors. These are inexpensive and widely available. See if you can change the script to get moisture data and include it in the CSV output. You could even send alerts when your plant needs some water. If you imported weather data from free public APIs such as OpenWeatherMap as well, you could see how changing conditions are affecting your plant!

More than plants

You can measure anything you want with this project, even growing youngsters, but why stop there? You’ve now had a taste of the power of OpenCV, which has many more capabilities than we’ve covered here. It is a powerhouse of computer vision that includes machine-learning capabilities that allow you to train your Pi to recognise objects, faces, and more. It’s a popular choice for robot builders and plays a major part in the autonomous challenges of Pi Wars. We can especially recommend pyimagesearch.com, which provided inspiration for this tutorial.

Click here to download the code from GitHub.