Projects

Home | Projects | Resume | Courses | Miscellaneous | Contact

• Custom Android kernel image on Motorola Droid (Love CMU for providing it!): Added resource management and admission control schemes to the Android kernel that provides timely, guaranteed and protected access to system resources. Also enhanced the Android kernel to implement two power management schemes, SysClock and PM-Clock. As an interface for the user-land to interact with this, added numerous system calls and sysfs files into Linux for making this possible.
  Platform: Motorola Droid
  Software: C, the (not-so-enormous) Linux kernel source code
  
  
• Gravel: A shared memory, pre-emptive, real-time operating system kernel for an ARM processor. It includes a system call handler, an interrupt handler, a timer driver, and a process management unit.
  Platform: ARM (a Gumstix motherboard).
  Software: C, ARM Assembly
  
  
• Tracking on DSPcam's: Tracking multiple objects across space and time on a distributed smart camera network.
  Advisor: Prof. Raj Rajkumar
  Platforms: Analog Devices Blackfin DSP, uClinux, DSPcam
  Software: C, OpenCV
  
  
• Privacy camera: Selective encryption of personally identifiable information on-board on a distributed smart camera network. We use wireless sensor motes to identify authorized users, and use user-settable privacy levels to mask/unmask features in videos recorded by the cameras.
  Platforms: Analog Devices Blackfin DSP, uClinux, DSPcam, FireFly Sensor Network
  Software: C, OpenCV
  
  
• De-identification of speech: Use of voice transformation to disguise speakers' identities while preserving intelligibility of speech. We found the 'best' transform from a set of transforms, such that the transformed speech is 'most different' form the source speech.
  Advisor: Prof. Alan W. Black
  Platform: Festvox, Festival
  Software: MATLAB
  Project Poster
  
  
• Texture Classification: Classification of materials based on their textured appearance using a single photograph captured under unknown viewpoint and illumination conditions. Tested using the CUReT database and the Amsterdam Library of Textures.
  Software: MATLAB
  Please send me an email if you want the source code
  
  
• Smart Gestures: A framework to enable front-camera equipped computers to respond to iPhone¹-like pinch and zoom gestures without the use of a touchscreen.
  Software: MATLAB