[ Home : Programs | libCVD | Hardware hacks | Publications | Teaching | TooN | Research ]

Videos

Tracking a desk lamp.

This video shows a desk lamp being tracked by following its apparent contour. The lamp is a good illustration of this tracking technique, since this is about the only useful feature of the lamp. The left hand frame shows the tracker in operation: the green lines are the edge-normal searches, and the blue line is placed where the tracker thinks the apparent contour should be. The top left frame shows a rendered version of the same view. The bottom left frame is a novel synthetic view from in front of the lamp.

See Rapid rendering of apparent contours of implicit surfaces for realtime tracking

First frame of the tracking video.
Tracking a desk lamp [MPEG2]
[2.8MB]

Tracking the inside of my lab.

This video shows the inside of my lab being tracked using a technique which robustly combines tracking information from a point based tracker and an edge based tracker. The strengths of the two systems can be seen in this video; large, unpredictable motions are tracked by the point baed tracker, and accumulated errors are removed by the edge based tracker.

See Fusing points and lines for high performance tracking.

First frame of lab tracking video
Tracking my lab [MPEG2]
[7.0MB]

Comparing unimodal and multimodal tracking.

This video illustrates the advantage of using a non-trivial method to combine the results from a point based and edge based tracker. If the point tracker is simply used to initialize the starting position of the edge tracker, then the overall system breaks on this sequence. If the multimodal nature of the posterior is taken in to account, then the edge tracker can effectively be tested for failure. Once failures can be detected, then they will not cause tracking to fail.

See Fusing points and lines for high performance tracking.

First frame of lab tracking video
Multimodal tracking [MPEG2]
[1.7MB]

Automated label placement

Videos of real-time automated label placement in an unstructured, cluttered environment. The videos illustrate object specific labels, object specific labels with directional placement constraints and a screen stabilized label.

See: Real-time Video Annotations for Augmented Reality.


Updated July 14th 2011, 11:54