"Designing a Consumer Panoramic Camcorder Using Embedded Vision," a Presentation from CENTR

Post on 16-Aug-2015

23 views 0 download

Tags:

Transcript of "Designing a Consumer Panoramic Camcorder Using Embedded Vision," a Presentation from CENTR

Embedded vision technology puts 360° HD video in the palm of your hand

Paul Alioshin - CTO

Our goal

CENTR’s goal is to transform the way people capture and curate the everyday photos and videos they share

Embedded vision technology provides substantial improvements in image quality and power efficiency, enabling low-cost, high-quality panoramic imaging

CENTR in action

How does it work?Click below for brief demo

Potential uses for 360°

• CENTR Camera seamlessly stitches videos together from four independent HD image sensors, all on the device

• Real-time, 4K-resolution panoramic video

• Low-profile design allows for everyday use

• Standard video format - easy sharing

• Footage can be streamed in real-time

• Partnership with Movidius allows unparalleled image processing

What sets CENTR apart?

• On-board key point detection and feature matching

• Error minimization across the entire 360° image: < 3 pixel error per seam

• Real-time distortion correction

• Dynamic color correction and exposure control across all four cameras

On board embedded vision

ISP pipeline power consumption

* based on four 5MP cameras downscaled to 720p/30fps ** based on 4K/30fps power estimates

CENTR*

CENTR* (w/ h.264)

Ambarella A9**

NVIDIA Tegra K1

0 1000 2000 3000 4000 5000

5,000

1,900

915

635

mW

ISP pipelineCamera x4Camera x4Camera x4RAW 5MP output (x4) Vignette Correction RAW Corrections Histogram &

Statistics

De-Bayer

De-Noise

Sharpen

Gamma Correct

Histogram & Stats (x4)Histogram & Stats

(x4)Histogram & Stats (x4)Histogram & Stats

(x4)

Choose 3 nearest WB

Global WB to avg. of the 3 Blend gains and add

to stitched photo

Set gains and exp. times (x4)Set gains and exp.

times (x4)Set gains and exp. times (x4)Set gains and exp.

times (x4)

Global Gain and WB Settings

Stitching Pipeline

Calibration Parameters

Distortion Parameters

Dynamic Stitching Map

Stitched Panoramic Image

Gain Map

Processed Image Data (x4)Processed Image Data (x4)Processed Image Data (x4)Processed Image Data (x4)

Under the hood• Image processing latency of <100ms

• Processes RAW image data for optimal quality

• Proprietary calibration algorithms optimize for 20 independent variables for the best image quality

• Average calibration error: <3 pixels per seam (based on US Army development program)

• Dynamic feedback loop between image sensors and CENTR algorithms account for changing lighting conditions and variable white-balance

• Real-time 360° video streaming/recording up to 60fps

• Up to 4K panoramic resolution

• 20 MP time-lapse panoramic photos

• Blue-tooth enabled mobile device controls

• CENTR image processing algorithms running on Movidius architecture

Enabling Virtual Presence

!

• Looking forward, CENTR’s real-time processing and hardware agnostic image processing algorithms allow rapid development of virtual presence systems for live events

• CENTR has rapidly developed 3, 4, and 5 camera systems for prior customers

Paul Alioshin - CTO !

paul@centrcam.com