Tools for Mobile Photography and Computer Vision

Explore computational photography and computer vision in this 51-minute conference talk from SIGGRAPH 2011. Gain insights into NVIDIA's research and development efforts in mobile, desktop, and server computing. Discover how these technologies drive future hardware and software roadmaps. Learn about Tegra 2's heterogeneous multi-core architecture, OpenCV functionality, and Android development for computer vision applications. Examine the challenges of mobile computational photography and the FCam architecture's solutions. Delve into practical examples of exposure metering, auto white balance, and contrast-based autofocus. Investigate advanced techniques like exposure fusion for HDR imaging and lucky imaging for hand-held long exposures. Understand the potential of computer vision in vehicle safety and new mobile applications. Access additional resources and development tools for implementing these technologies on NVIDIA platforms.

Intro
NVIDIA Mobile Computer Vision
Computer Vision = New Applications
Computer Vision = Vehicle Safety
Mobile Computing on a Tear
Tegra 2 - Heterogeneous Multi-core
ARM is Pervasive and Open
TEGRA ROADMAP
OpenCV Functionality Overview Image processing
OpenCV for CUDA
Current OpenCV Cuda Functionality
OpenCV for Android
Tegra Android Development Pack
Android Apps Java Applications
OpenCV supports Java and Native
OpenCV Java API Example
The Embedded Vision Alliance
Computer Vision vs. Photography
Some computational photography apps
Mobile Computational Photography
MCP Problem 1: Lack of control & accesso
MCP Problem 2: Wrong model
The FCam Architecture
Sensor
Image Signal Processor (ISP)
Devices
Implementations
Tegra developer board
FCam: Open Source Project
Exposure metering examples
Auto White Balance
Estimating the color temperature
Contrast-based Passive Autofocus
Exposure Fusion: Simplified HDR
Implemented on N900
Lucky Imaging: Hand-held long exposures
First assignment: Autofocus - 1 week
Remote flash over Bluetooth
Interactive Photomontage
Kal-El Reference Design
FCam on Tegra: Baby steps