By exploiting tracking information in a semi-supervised learning framework, a classifier trained with three hand-labeled training tracks can produce accuracy comparable to the fully-supervised equivalent.
Quantitative results on the track classification sub-problem
In the spirit of divide-and-conquer, this test set does not include segmentation and tracking errors.
Qualitative results on the full object recognition problem
Object recognition results computed using laser range finder data are visualized in video. (Camera data is not used in object recognition; it just makes it easier for humans to look at.)
The classifier used to generate these results was trained using 3 hand-labeled training tracks of each object class plus a large quantity of unlabeled data. Gray outlines are objects that were tracked in the laser and classified as neither pedestrian, bicyclist, nor car.
Segmentation and tracking errors (beyond the scope of this paper) are the largest sources of error; track classification is comparatively reliable.
Page generated 2013-08-23 15:34:15 PDT, by jemdoc.