Description
The aim of this project is to assess the effectiveness of SIFT features for object recognition.
To accomplish this aim, select a traffic sign (e.g., see this link) and using your mobile
device, take ~10 images of outdoor scenes that include the selected sign. Make sure to
include various sizes and poses of the sign, including perhaps occlusions. Use the SIFT
implementation available at VLFeat.org to investigate the matching capabilities of SIFT
under different geometric and photometric variations.
You may work in groups of 2 or 3 and upload the following two items to Blackboard
as a single ZIP file.
• A brief report that summarizes the results of the recognition task. This may be
presented as a table that includes the type of deformation and the accuracy of the
matches.
• Matlab code that, when executed, takes each of the test images one at a time (with
pauses in-between) and displays the matched SIFT points between the image and
the reference traffic sign. Be sure to include the test images and the reference sign
in the ZIP file.
Note: This project will count as two quiz grades.
Deadline: Midnight on Nov. 27th
