Persönlicher Status und Werkzeuge

Home Research Cop: Cognitive Perception

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
research:cop [2011/08/01 15:23]
tenorth
research:cop [2011/08/01 15:27] (current)
tenorth
Line 66: Line 66:
 shared databases available over the Internet. ​ shared databases available over the Internet. ​
  
-{{ :​research:​cop:​cop10.png?​nolink&​200|}} 
  
 This approach, This approach,
Line 73: Line 72:
 like wrong annotation due to the open nature of the database, like wrong annotation due to the open nature of the database,
 or overwhelming amounts of data (many 3D models) or the lack or overwhelming amounts of data (many 3D models) or the lack
-of relevant data (no models matching a specified label). To solve+of relevant data (no models matching a specified label). ​ 
 + 
 +{{ :​research:​cop:​cop10.jpeg?​nolink&​200|}} 
 +To solve
 those problems we propose the following: First, we present an those problems we propose the following: First, we present an
 outlier/​inlier classification method for reducing the number of outlier/​inlier classification method for reducing the number of
Line 83: Line 85:
 our knowledge of the environment. ​ our knowledge of the environment. ​
  
-{{ :​research:​cop:​cop11.png?​nolink&​200|}} +===Eye Watch You=== 
- +{{ :​research:​cop:​cop12.png?​nolink&​200|}} 
- +Inside-out action analysis shows new aspects in observation 
 +of daily life activities. Especially, manipulation tasks 
 +can be understood easier by an observer seeing the view 
 +of the acting person. The gaze provides early information 
 +about focus of the attention and critical points during an 
 +action. Any action is prepared by observing the area of interest 
 +before manipulating it.  
 +{{ :​research:​cop:​cop13.png?​nolink&​200|}} 
 +Additionally,​ humans mostly 
 +keep their focus on their hands while grasping an object. 
 +This fact allows us to automatically extract hand movements 
 +out of images of a gaze directed camera in relation to the 
 +acting person and the position in the world, that can be 
 +tracked in parallel, given an adequate world model. We propose 
 +in this work a novel hybrid 2D-3D method for markerless 
 +{{ :​research:​cop:​cop14.png?​nolink&​200|}} 
 +hand tracking under the difficult conditions such a 
 +gaze directed camera imposes:  
 +gaze changes with very high 
 +speed from one scene to another, with the effect of blurred 
 +frames and and little scene overlaps. So, the major problem 
 +that we address is the case of losing track of a hand. Given 
 +such an event the image is scanned for appearing hands until 
 +a hand’s present is validated over several frames. This 
 +validation is performed using a simple 2d hand model that 
 +equally serves initializing an more exact 3d hand model that 
 +is used to track the hand till the next track loss. Based 
 +on several distance measurement,​ we decide for an optimal 
 +hand configuration,​ as well in 2D as in 3D. The distance 
 +measurements are based on skin color, edges and edge directions. 
 +Our investigations are targeting as well on action 
 +analyzing in general as for generating a handbook of hand 
 +trajectories,​ grasping points and gaze directions in relation 
 +to the object of interest and possibly obstacles for controlling 
 +a service robot.
  
 ==== Acknowledgements ==== ==== Acknowledgements ====
Last edited 01.08.2011 15:27 by tenorth