Practical 3-D Object Detection Using Category and Instance-Level Appearance Models

TitlePractical 3-D Object Detection Using Category and Instance-Level Appearance Models
Publication TypeConference Paper
Year of Publication2011
AuthorsSaenko, K., Karayev S., Jia Y., Shyr A., Janoch A., Long J., Fritz M., & Darrell T.
Other Numbers3210

Effective robotic interaction with household objectsrequires the ability to recognize both object instances andobject categories. The former are often characterized by locallydiscriminative texture cues (e.g., instances with prominentbrand names and logos), and the latter by salient globalshape properties (plates, bowls, pots). We describe experimentswith both types of cues, combining a template-and-deformablepartsdetector to capture overall shape properties with alocal feature Naive-Bayes nearest neighbor model to capturelocal texture properties. We base our implementation on therecently introduced Kinect sensor, which provides reliable depthestimates of indoor scenes. Depth cues provide segmentationand size constraints to our method. Depth affinity is used tomodify the appearance term in a segmentation-based proposalstep, and size priors are imposed on object classes to prunefalse positives. We address the complexity of scanning windowHOG search using multi-class pruning schemes, first applyinga generic object detection scheme to prune unlikely windows,and then focusing only on the most likely class per remainingwindow. Our method is able to handle relatively cluttered scenesinvolving multiple objects with varying levels of surface texture,and can efficiently employ multi-class scanning window search.

Bibliographic Notes

Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Algarve, Portugal, pp. 793-800

Abbreviated Authors

K. Saenko, S. Karayev, Y. Jia, A. Shyr, A. Janoch, J. Long, M. Fritz, and T. Darrell

ICSI Research Group


ICSI Publication Type

Article in conference proceedings