Computer Vision-based application for robotics rely on extracting meaningful features from images which can be used to perform many tasks like place recognition, place categorization, tracking, Visual Odometry, SLAM, object detection, object recognition, etc. All these tasks rely on a set of key-points and descriptors that describe them. Each of the features/key-points are associated with some sort of key signatures (called descriptors of those key points). Many existing methods for detection/description of these features exists, each descriptor has certain properties that are suitable for certain vision applications. Some of the existing methods include SIFT, SURF, BRISK, FAST, ORB,AKAZE, LATCH, LSD/BLD (lines), etc. Some of the detectors already provide descriptors but some of them do not, so different combinations of descriptors can be used. In this project a GUI application is built which evaluates these descriptors/key-points. The app also provides a means to benchmark these features for certain vision tasks like Visual Odometry, Tracking and Place Recognition. A second mini project of extending the MRPT ROS package to playback MRPT Datasets in ROS RViz is also implemented.

Student

Raghavender Sahdev

Mentors

  • Francisco-Angel Moreno
  • Feroze Naina
close

2017