This was a 2 semester long team project which I took up at Carnegie Mellon's Robotics Institute. We were on a mission to build the first fully autonomous quadrotor with a docking station which powered it for repeat flights.
Our vision was to make a system that could collect aerial data for farmer but at the same time be autonomous enough so that the farmer is oblovious to it after the initial setup. Basically we wanted to take the drone out of the equation! One of our team mates spun off his startup from this project.
I implemented Inertial Measurement Unit (IMU) based Image Stabilization for global shutter cameras. This was a project at Tonbo Imaging, Bangalore, the company I worked for before graduate school.
The idea was to estimate motion using an IMU and correcting the jittery video feed from a vibrating camera in real time. I used the IMU values to calculate the registration matrix between consecutive frames and then used the registration matrix along with a Kalman Filter to stabilize the video feed.
Check out the results of this project here and here.
Lost Child Google Glass and Android App
I implemented an Android app that helped find a lost child in a crowd using Google Glass stream as input and deep learning based face matching. A demo and presentation of it can be seen here. This was for a class project in the Mobile and Pervasive Computing class at CMU.
Face Alignment iOS App
I implemented an iOS app that did face alignment and landmark detection on an iPhone in near real time. A demo and presentation of it can be seen here. This was for a class project in the Computer Vision Mobile Apps class at CMU.
Pulse Rate Estimation using Facial Video
I implemented a C++ module that used the video feed of a person's face to estimate their pulse rate. I worked on this at Neuroinformatics & Cognitive Robotics Lab at TU Ilmenau, Germany. This was for a robot that assisted residents at home for the aged. Its major application was to assess the health of the person when they exercise.