Development of an Algorithm for Dividing Video Stimuli into Areas of Intesests (AOI) for Synamic Eye-Tracking Data Pre-processing

This thesis is part of a larger study on analyzing Dynamic (Interactive) Stim- uli Eye-Tracking Data of students with the end goal of finding differences between novice and expert programmers when it comes to program comprehension and code debugging. For the grouping of video frames into scenes based on the desktop en- vironment, a Video Frame Scene Grouping Algorithm was developed which is able to group different frames of a video into different scenes based on the desktop envi- ronment using Structural Similarity Index (SSIM). For the estimation of potential Areas of Interests (AOIs) and their boundaries, an Automatic AOI Bounding Boxes Estimation Algorithm which uses Image Sharpening, Histogram Equalization, K- Means, Bilateral Filtering, and SLIC was created to determine potential Areas of Interests. These two algorithms are then brought together into one algorithm that utilizes multiprocessing to perform Video Frame Scene Grouping Algorithm simul- taneously on multiple videos and Watchdog and threading to perform Automatic AOI Bounding Boxes Estimation Algorithm simultaneously on multiple frames of a video across scenes. Qualitative and Quantitative analyses show that while both algorithms are very effective in performing their intended function, there are in- deed rooms for improvement most especially when it comes to the runtime, and as well the accuracy of the Automatic AOI Bounding Boxes Estimation Algorithm