Week 11 Report

Over the term, Dr. Stonedahl and I have worked on the videos of the chicks to find a way to accurately track their movements and create a model of them. So far, we have accomplished the following goals:

• ·       We effectively remove the background of the video by using Gaussian blur and threshold to convert the video into black-and-white. We further use erosion to reduce the area of the chicks so that it decreases the occasions when the chicks move next to one another and they merge in the process. We also use the contour tool to mark the center of the chicks (the blue dot) because it makes it easier to store the coordinates of the current position of each chick further on.

  

  Background Removal 

• ·       Learning from the Whirligig Beetle Project, we also successfully create a multiple frame tracking function. We create a Chick class where it stores the current location of one chick. Then we use 3 lists of the Chick object to keep track of all the chicks: one to store the chicks that currently being tracked, one to store the chicks that are no longer being tracked (after being lost for 5 frames) and one to store the chicks that were tracked for at least 200 frames. For differentiation, we mark each chick with a unique color pointer and a unique string of number and character. We then draw a graph of the movements of the chicks based on the active list that we have.

  

  Multiple Frame Tracking

• ·       To test the accuracy of our tracking method, we create a function to get the position of each chick manually (by manually clicking every 1 second to get the coordinate of the current position of a chick and put it into a list, then storing that list as a .csv file). So far, we have got 9 files of data of the 9 chicks (the 3-chicks data are from frame 4110 to frame 24990, the 6-chicks data are from frame 39420 to frame 48480). We then draw a graph to illustrate the movement pattern of each chick.

  

  Manual Tracking

The following list is what I think could be further worked on and improved:

• ·       The current tracking method can still be improved. At the moment, it does not do a good job of detecting when a group of chicks “merge” together. As a result, it is only accurate when the chicks are a bit away from one another.

  

  The chicks merge

• ·       We previously tried to track the chicks by using Meanshift (a function within the OpenCV package). However, it only works for the first few frames and then fails to track the chick. We have not figured out the reason it fails but the potential is there. It needs further work on.

  

  Meanshift Tracking (the square)

• ·       A model of the chicks using NetLogo.

Below is a video about the tracking method (the red dot is the accurate position of the chicks that was done manually)