|Surveyor Robotics Journal|
Sun, 02 Nov 2008
one more pass at color segmentation + edge detection
Code for transferring data between processors via SPI bus on the SVS (stereo vision system) is in place, so now we need to decide what data to transfer for the purpose of building basic depth maps. There is no question that we will eventually need to move toward more robust feature detectors such as SIFT or SURF, but we should be able to get useful results from the simple combination of an edge detector and basic color segmentation.
The previous robotics journal post showed an edge detector combined with a 16-color segmentation based on just 4 levels of U and 4 levels of V, but no Y component. Shown below is 16-color segmentation based on the 'vz1' filter with 4 levels of Y, 2 levels of U and 2 levels of V. Combined with the edge detector, this approach seems to provide better feature differentiation than the UV-only approach, so that is the approach we will use as a basis for correlating between left and right views, though it is easy enough to change to other segmentation schemes.
displayed using SRV1Test.java