Application: track-video-features

1. Description

The program track-video-features demonstrates live detection and tracking of features in a video stream.

To invoke it, just type from the console:

$ track-video-features

and a GUI dialog will ask you to pick the video source (among video files, rawlogs, live firewire or USB cameras, a Kinect sensor, etc.). Alternatively, use:

track−video−features <DATASET.rawlog> track-video-features <video_file.{avi/mpg/mp4/flv}>

to directly run it with a robotic dataset or a video file.

2. Source code

The program is located at <MRPT>/apps/track-video-features/: https://raw.github.com/MRPT/mrpt/master/apps/track-video-features/track-video-feats_main.cpp

See this tutorial on the related data structures and classes.

3. Demo videos

1. A short sequence in a typical office scenario: Test with a sample video sequence published by Andrew Davison among his MonoSLAM works.

2. An unmodified music video: This kind of videos are really challenging for computer-vision methods for the quick sequence changes, and for blurred and unfocused shots. Notice how features are hardly detected in those blurred images; the reason is that at present (sept-2011) only the first image pyramid is searched for FAST features in the program.