Linas Kondrackis
  • Portfolio

Click on the images to enlarge.

Multiple 'robot' tracking in images

Computer Vision • Matlab

This task required us to track three colored ‘robots’ in a sequence of images. Images had three different background types and had to work with an unknown forth one, only revealed during assessment.

To tackle the task, we subtracted the background and found morphological operations on the image to reveal the triangles on the top. Shapes were verified to be triangles using a ratio of perimeter and area. Triangle corners were then found - for this K-means clustering was used due to a large number of datapoints at the ends of the triangles. Color reading was then taken just behind the triangle, distinguishing the ‘robot’.

In the end, the solution was robust and managed to track the ‘robots’ successfully (the failure cases shown occurred during development). The only issue we did not manage to fix was an inefficiency in the plotting, gradually slowing down the program.

Report: Google Drive

If you would like to take a look at the code, please contact me.



  • Category: Projects
  • Date: January 2016 - February 2016
  • Collaborators: Dalius Grazinskis

Perfectly detected and classified 'robots'. The triangles are not noisy at all, corners are detected perfectly and 'robots' are each assigned a unique color value, most likely matching with the actual color of the 'robot'.

Insufficiently detailed detection. Either color detection thresholds are not fine enough, or dilation is too strong.

Due to noise, incomplete triangles are formed, preventing corner detection. Therefore, these shapes cannot be classified as 'robots'. Furthermore, noise below the 'robot' is visible, hinting the lack of erosion. A difficult situation as the two upper triangles are too eroded.

The three 'robots' on the first background. You can see that our methods were successful in plotting their motion path.

The three 'robots' on the second background. You can see that our methods were successful in plotting their motion path.

The three 'robots' on the third, the most difficult, background. You can see the blue 'robot' misclassified as red and the green one having his direction incorrect. The red 'robot' is not recognized at all.