---

Table of Contents

1. My Setup
2. Problem
3. Solution
4. Takeaways

I track outdoor activities via Obsidian and show them here. Since I changed my EDC to include a notebook here and I thought it would feel nice to have a paper example too - as paper feels, and is, real.

The value of the paper is I can take 2-3 minutes in the morning, sit down, look out the window, scratch off the day, provide any notes, mark if there was an activity, and finally make a tally mark. It is calming, not that I’m not calm, but it adds some nice regiment to my day.

I now have the problem of my digital and paper calanders of activities being out of sync.

I care because they are supposed to match! As in, they don’t have different data.

How could I take a picture of my calendar and get insights from the picture to know if I’m in sync or not?

OpenCV seemed like the tool for the job, to evaluate images.

I tried looking for crossed out numbers at first, that didn’t make sense.

I tried counting tally marks, that was pretty inconsistent.

I landed on: could OpenCV detect highlighter marks and count them? For days with activities I put a highlighter within that day.

What I need to do:

1. Upload an image
2. Process it against OpenCV
3. Persist the month and count
4. Report to myself when they are out of sync

I already have Kestra stood up, thought I might as well use it as it can take parameters, like images, which is pretty cool!

Disclaimer: I did use ChatGPT to get things rolling.

Here is my workflow in Kestra:

id: uploadCalendar
namespace: dev
labels:
  - key: type
    value: upload
inputs:
  - id: my-file
    type: FILE
tasks:
  - id: go
    type: io.kestra.plugin.scripts.python.Script
    taskRunner:
      type: io.kestra.plugin.scripts.runner.docker.Docker
    containerImage: jupyter/scipy-notebook
    beforeCommands:
      - pip install opencv-python
    warningOnStdErr: false
    inputFiles:
      file.tmp: "{{ inputs['my-file'] }}"
    script: |
      import cv2
      import numpy as np
      import requests

      # Load the image
      image = cv2.imread("file.tmp")

      # Convert to HSV color space
      hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)

      # Define color range for green highlighter (adjust if needed)
      lower_green = np.array([40, 50, 50])   # Lower bound of green
      upper_green = np.array([80, 255, 255]) # Upper bound of green

      # Create a mask for the green highlights
      mask = cv2.inRange(hsv, lower_green, upper_green)

      # Find contours in the mask
      contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

      # Draw contours and count them
      highlight_count = 0
      for contour in contours:
          area = cv2.contourArea(contour)
          if area > 500:  # Filter small noise (adjust as needed)
              highlight_count += 1
              cv2.drawContours(image, [contour], -1, (0, 0, 255), 2)  # Draw detected highlights in red

      # Display results
      print(f"Detected Green Highlighter Strips: {highlight_count}")
      url = "http://192.168.86.100:3333/api/dump/calendar_processed_v1?token={{ secret('KESTRA_WEB_API_TOKEN') }}"
      print(url)
      data = {'highlight_count': highlight_count}
      r = requests.post(url, json = data)
      print(f"Response: {r.reason}")

1. Pretty awesome you can have file inputs in Kestra, especially images
2. It was a pain to get OpenCV installed with all its dependencies, so I finally found a container that had what was needed
3. The persistence is arbitrary
4. I couldn’t get secrets to work. I’m showing one being used, but I have my api key hardcoded in the real yaml. It was pretty frustrating
5. This all works from my mobile phone! I just navigate to the web page I have bookmarked

To reduce friction, I can choose execute the workflow, choose the file input, and choose take image. So no need to find an image from my gallery, just take it ad-hoc.

---

Back to Top