FEATURE SELECTION AND TRACKING

---

This laboratory exercise will guide you through exploring the feature detection and tracking algorithms described in Chapter 11 of "An Introduction to 3D Vision", by Y. Ma, S. Soatto, J. Kosecka and S. Sastry (MASKS).

The goal of this session is to experiment with feature detection and tracking algorithms. In particular, the reader should gain intuition on the role of various design parameters such as the size of the windows for selection and tracking, the SSD criterion and thresholds. Here is a list of MATLAB files that you will need to download and save in a local directory:

• SelectFeature.m a Matlab function that takes as input an image and returns the location of a maximum number of point feature
• track.m a Matlab function that takes as input a number of images taken at adjacent instants in time (and optionally the initial location of a set of feature points), and returns the position of the features in each subsequent image
• ComputeQuality.m , GridQuality.m , LocalMax.m , downsample.m , boxes1.bmp
  
• a sequence of images

Alternatively, download a g-zipped, tar file with all the necessary files and data.

Launch the function DetectionDemo to see how point features and corners can be detected on images. Then open the file DetectionDemo.m with your favourite editor, and follow it step-by-step to detect features and corners in the lab scene.

Launch the function TrackDemo to see how point features can be tracked from frame to frame in a video sequence. Then open the function TrackDemo.m with your favourite editor, and follow it step by step to track point features in a sequence.

• After an initial set of putative correspondences is established, use RANSAC to refine the localization of features in each frame.

---

An Invitation to 3D Vision Y. Ma, S. Soatto, J. Kosecka, S. Sastry THE CODE ON THIS PAGE IS DISTRIBUTED FREE FOR NON-COMMERCIAL USE. Copyright (c) MASKS, 2003.

---