The motivation of this project is to summarize lengthy video footage (like CCTV cameras) to provide a video synopsis which would simultaneously convey various events that take place with time stamps.
Using a combination of background subtraction, connected components analysis and kalman filters we were able to design and implement a sysnopsis video algorithm. You can watch both the original video and the synopsis below.
Using Neural networks we attempted to predict the word a person is thinking by looking at their brain’s fMRI scans. We build a classifier for cross subject decoding using a Convolutional Neural Network.Spring 2016
Currently working on a sensor based network to detect ball-player contact in football (soccer) to be able to create a internet of things system to detect offside calls.
Built a classifier to predict motor actions using fMRI data of the test subjects brain. The unsupervised Classifier predicted missing
fMRI data used Regularized EM to get an RMS error of 0.45. The supervised classifier predicted if a person was pressing a button accurately when prompted to with 69% accuracy.
Implemented CSP Algorithm (Constraint Satisfaction Problem) to automate scheduling of appointments in simulated hospital data. Using CSP improved appointment cancellation rates and provided a solution for no shows.
Implemented A star search to find the shortest path between any two intersections in Pittsburgh using real data exported from OpenStreetMap
A sorting game for kids where they have to sort three different shapes into 3 different bins in 60 seconds. The system will automatically keep track if the correct shapes have been sorted. The project was displayed at the Pittsburgh children’s museum. You can watch the system we built in action below.Spring 2015
Designed and Implemented a pair of swarm robots which worked together to transport an object from point A to B.Spring 2014