Research:
As part of my Master's program at IISc, I worked on Foreground-Background Segmentation and Object Tracking for Video Surveillance Applications. I used
the GMM-based background modelling principles for Segmentation purpose. For
training the background model, I used certain fast algorithms which drastically cut down the training time. Incremental changes to the segmentation algorithm was also made.
I used the Level Set methology for both Tracking and Shape-Contour Extraction of foreground objects. Also issues reagarding the stability and convergence of Level Set algorithms were delved upon and I finally came up with a convergence and a stability criteia.
Demonstrations:
I have trained my background model using an input video sequence
consisting of only the background.This trained model is used for
classification of foreground objects. The segmetation result
has a lot of nosise in it, but the foreground objects are clearly
discernable. After post processing (for noise removal) and connected
component analysis, the foreground object are segmented from the background.
Also for perfect Shape-Contour Extraction and Tracking, I have tried
the modifications of the Level Set methodology, now very popular in the
Computer Vision arena. The curve evolution principle for shape-contour extraction, using the Eulerian formulation, gives satisfactory results. The Tracking results using Level Set implementations are also very promising.
Publications:
(1) Submitted a paper, titled "Stability and Convergence of Level Set Methods in Computer Vision", to the Pattern Reocognition Letters journal.