Ant Tracker

• [Oct 2020 – Present]

Code: https://github.com/ec888/motion-tracker

I’ve always been fascinated by ants. I spent hours and hours staring at them as a kid, with infinite questions

1. Experimental questions

This year, ants invaded my house, which was a great opportunity to start studying them formally. Having done image processing-related work over the summer, I knew a pretty solid path to this project was within reach. It’s challenging to distill all the abstract ideas and questions floating around in my head into experimentable questions, but I ultimately decided on these four primary goals to address:  

1. Can we quantify/graph ant communication?
2. How do ant trails form? How do they respond when their normal foraging paths are blocked?
3. If we can AI to simulate each individual ant, will the resulting collective intelligence be the most robust and efficient? Or can we improve on it? Why does emergent intelligence form? 
4. What is the fractal dimension of an ant’s foraging path? Can “confusion” be robustly quantified with fractal dimension? 

2. Setup & Code

This image shows the simple setup I used to record ant videos. The black rectangle is a magnet ring (I observed ants walking in less directed patterns when walking on the ring, so wanted to quantify it). Then, I performed other analyses as well with other materials instead of magnets: citrus, toothpaste, spider silk.

The first step is to build a tool to analyze ants video and collect ants path data. There is no widely used open source tools for this purpose. I forked and modified this project in Github, and was able to collect path data. I plan to use it for future research on ant behaviors. 

Luckily, Hunt et. al (2018) did the heavy-lifting for this so I didn’t have to repeat it:

“The paths of ants were traced out manually from the recorded videos by following a navigating ant with the computer mouse cursor …”

Challenges: 

1. The code was written in Python 2. I ran it on Python 3 after a few syntax fixes. However, it did not work as all the ant paths were dropped. With internet search, I found the culprit: in Python 3, the map() function return an “map” object, which is an iterator instead of a real list object. Once you iterated through it, it changed to empty and subsequent use would yield empty list. I followed a python migration guide to convert the map object into a real list and solved the problem.  
2. The code uses OpenCV and functional programming style, which I was not originally familiar with. I studied each line of the code and added documentation. 

Next blog post will probably be about results. To be continued!