I developed a novel six degree of freedom wireless pose capture system. This system makes use of an Intel RealSense camera for capturing the position of a vision target, and a BNO055 sensor for obtaining the orientation of the vision target. This pose can be used to move objects in a gazebo enviornment, or as a human-robot interface. The generated software, firmware, mechanical and electrical hardware designs can be found in the GitHub repository here. For more information on the project, please see the project page here.
In some of my master's research efforts, I developed a salient region of interest motion detection algorithm. For more information about the development effort, please have a look at the project page here. The source code generated for this project can be found in the GitHub repository here.
I was asked by the RIT space exploration club to develop a piece of software to detect the horizon in images aquired on high altitude balloons. This code was run on a raspberry pi flying at altitudes of 60,000+ feet! Here is a good horizon detection, with the angle offset computed in a counter-clockwise manner, printed in degrees. Have a look at the GitHub Repo.
I have compiled a tutorial on how to train Haar-based cascade classifiers using the opencv haartraining app. My tutorial as well as some example usage can be found on the Haar cascade training tutorial page.
I have developed an app that can test these Haar-based cascade files on the go. Paste the URL of a plain text XML file into the text box and hit the button! once the link text disappears you can the click on "my_classifier" from the classifier selection menu. This menu is obtained by clicking on the three dots at the bottom of the screen. This app is available on android devices through the Google play store here.
I have been working on an augmented reality vision system that will allow one to see different 3D projections using OpenCV. depending on the particular icon in the corner of the checkerboard target, the corresponding 3D image will come up. The icons are recognized by a Haar classifier. The code can be found on my Github. See below for a video of it in action.