Leveraging Geometric Constraints for Improved Structure from Motion
We present a multistage approach for SFM reconstruction of a single component that breaks the sequential nature of the incremental SFM methods. Our approach begins with quickly building a coarse 3D model using only a fraction of features from given images. The coarse model is then enriched by localizing remaining images and matching and triangulating remaining features in subsequent stages that are made efficient and highly parallel by leveraging the geometry of the coarse model.
Project Webpage |
Video Collections in Panoramic ContextsWe create a video-collections+context interface by embedding videos into a panorama and build a spatio-temporal index for fast exploration of the video collection. We study the effect of our system for spatio-temporal localization tasks, and find significant improvements to accuracy and time in visual search tasks compared to existing systems.
Proeject Webpage |
Interactive Video Navigation and Manipulation We enable simple and meaningful interaction for object-centric navigation and manipulation of long shot videos, by using the 3D (space-time) object trajectories as basic interaction elements and define several object and camera operations as simple and intuitive curve manipulations.
|
Automatic Photo Ehnancement using Facial Expression AnalysisGiven a set of photographs of the same group of people, our algorithm uses facial analysis to determine a goodness score for each face instance in those photos and creates a best composite by replacing any low-scoring faces with high-scoring faces of the same person from other photos, using seamless composition.
|
GPU-accelerated Genetic AlgorithmsImplementation of Genetic Algorithms on NVIDIA Tesla GPU, using CUDA Programming Model. A new gene-parallel approach was proposed and implemented and a very high speed up of up to 1500x was achieved.
|
Content based Video BrowsingInstead of timeline based browsing, content based scene browsing of the movie LOTR was enabled using various computer vision techniques for shot segmentation, scene classication, actor recognition etc.
|
Region of Interest zooming in VideoVideo being a temporal entity, enlarging a spatially fixed window does not serve the purpose of observing a region of interest over time. A visual tracking method based on SIFT and color features was employed to track the Region of Interest over time and shift the zoom region accordingly.
|
Study of Non-parametric Texture Synthesis Methods As a part of thisproject we studied various approaches for non-parametric texture synthesis and implemented three different approaches - pixel based, pyramid based and patch based synthesis.
|
Voice Conversion SystemIn this project we proposed and implemented two approaches for converting a source speaker’s voice to a target speaker using source-filter model of human speech production system. Also created a demo for interactively morphing voices by modifying parameters.
|
Portable Media PlayerDeveloped a Portable Multimedia Player using an 8-bit general purpose microcontroller ATmega128 and peripherals like touch screen, color LCD, MMC Card and MP3 playback chip. Attempted to develop a similar player using a 32-bit Application Processor AVR AP7000.
|