Papers/Machine learning

Feature Extraction from Depth Maps for Object Recognition

tomato13 2015. 5. 13. 18:48

This paper suggests features for depth image. It said when using HOG with depth images, much improved performance can be made which out-performs standard HOG. 


The below is the impressive statements. 


Two baseline features were used for comparison: standard HOG over the calculated RGB grayscale image itself and HOG using the depth map as a grayscale image. Both of these were intended as baselines, and HOG over the RGB grascale is the only feature in this paper that directly incorporates numerical color data.