Stereo Vision Algorithm Demo

Hawk Wang (yourwanghao@gmail.com)
2015-10-24

Introduction

This demo shows the stereo correspondance algorithm comparison between my algorithm and a competitors'.

The algorithms are running on a dual camera phone that has one 13M pixels camera, and one 2M pixels camera. You can get more information on my shiny app at https://yourwanghao.shinyapps.io/StereoVisionDemo

The baseline between the two cameras is 20mm, and below is the camera module:

Main Image and Sub Image

I use the two cameras to take photo at the same time so that I can use stereo vision algorithms to generate disparity maps. Note: Sub image is 90 degree to main image.

Disparity Map Comparison

Mine	Competitor’s

Performance Comparison

plot of chunk unnamed-chunk-1

As you can see from left figure, we test the depth generation for 10 times. And the result shows that mean time for competitor lib is 5.359 seconds, while for My lib it is 2.097 seconds. Apparently, my lib is much faster than the competitor's lib.

Note: The cell phone is based on QCOM MSM8939 platform, disparity map is 896 x 672 pixels