|
|||
|
File
Formats and Directory Structure Image Sets: |
Bicycle
Number Of Images: 151 Image Size: 2048 x 1536 Comments: These images provide a benchmark data
set for the challenging problem of recognizing articulated wiry objects. The target object, the bicycle, is
viewed at 3 different poses: with the long axis of the bicycle making an
angle of 90 degrees, 45 degrees, and 0 degrees (shown in the example image)
with the image plane. Moreover,
at each bicycle pose, the handlebars rotate to one of 3 different
articulations with respect to the rest of the bicycle: the front wheel makes
an angle of 0 degrees, 45 degrees, and 90 degrees with the long axis of the
bicycle. The camera height for these
images was about 1.9 m; the objects are roughly 4-5 m away from the camera;
and the camera rotates a total of about 60 degrees w.r.t. the bike. The camera shifts from side to side
between each image. Once every
3-5 images the pose and articulation of the bike is modified, and the clutter
objects are shuffled. Besides
articulation, other challenging aspects of this data set are: the unstable
nature of edges extracted from the plants, the sometimes large number of
edges detected among the bricks in the background, and the gradual change in
lighting over the course of the data set (caused by sunset). Vista Parameters: (sigma, lowThreshold,
highThreshold,minimumEdgeLength, accuracy, granularity, magnitude) =
(3,10,50,50,2.0,4.0,10.0) Download: (copyright) [bike.tar.gz , 108 MB] |
||
WORD: Wiry
Object Recognition Database
