Information Technology Reference
In-Depth Information
because without further clues, we identify the person's location as the object
location he/she is interacting with. However, in practice the object is usually a
short distance away from the person. But still the results are helpful to indicate
the objects and their location in the environment. Note that in front of the sofa,
there is a region misclassified as TV. This is because the algorithm generously
considers all grids along the gaze direction are possible TV locations. While some
grids are identified as other objects if other activities happened there, some (like
this region) have not been attended by the person. Therefore they retained the
hypothesis of being TV. This also explains the low precision of TV. Further
observations on this region would help resolve its identity.
7Conluon
In this paper we described a system to recognize objects in the smart home
environment with camera network. The objects are recognized through object-
activity interactions. A hierarchical activity recognition process is described,
which provides fine-grained activities. The object-activity relationship is en-
coded in the knowledge base of MLN. We described the details of the knowledge
base and inference process. Experiments are shown in the AIR lab smart home
environment. Future work includes combining the position-based object type
inference with image segmentation for better localization of objects.
References
1. Richardson, M., Domingos, P.: Markov logic networks. Machine Learning 62,
107-136 (2006)
2. Tran, S.D., Davis, L.S.: Event modeling and recognition using markov logic net-
works. In: Forsyth, D., Torr, P., Zisserman, A. (eds.) ECCV 2008, Part II. LNCS,
vol. 5303, pp. 610-623. Springer, Heidelberg (2008)
3. Pinz, A.: Object Categorization. Foundations and Trends in Computer Graphics
and Vision by Now Publishers (2006)
4. Oliva, A., Torralba, A.: The role of context in object recognition. Trends in Cog-
nitive Sciences 11, 520-527 (2007)
5. Heitz, G., Koller, D.: Learning spatial context: Using stuff to find things. In:
Forsyth, D., Torr, P., Zisserman, A. (eds.) ECCV 2008, Part I. LNCS, vol. 5302,
pp. 30-43. Springer, Heidelberg (2008)
6. Peursum, P., West, G., Venkatesh, S.: Combining image regions and human activity
for indirect object recognition in indoor wide-angle views. In: Proceedings of the
Tenth IEEE International Conference on Computer Vision (ICCV 2005), vol. 1,
pp. 82-89 (2005)
7. Gupta, A., Davis, L.: Objects in action: An approach for combining action un-
derstanding and object perception. In: IEEE Conference on Computer Vision and
Pattern Recognition, pp. 1-8 (2007)
8. Oliver, N., Garg, A., Horvitz, E.: Layered representations for learning and infer-
ring oce activity from multiple sensory channels. Computer Vision and Image
Understanding 96, 163-180 (2004)
Search WWH ::
Custom Search