Flexible Hybrid Stereo Matching
Jingzhou Huang, Changyu Diao · Advances in intelligent systems research/Advances in Intelligent Systems Research · 2015
We present a hybrid stereo matching algorithm, which decomposes the stereo matching process into a number of subtasks and deals with them using different methods.Our algorithm consists of three components: 1) segment the reference image into regions and regard each region as the basic task unit; 2) extract and match feature points of the stereo image pair; 3) determine the type of each task unit and handle it with a certain method.Tasks can run on CPU and GPU simultaneously, so different computing resources achieve a real cooperation.We test our algorithm on a PC with single CPU and single GPU, and results demonstrate its effectiveness.