CAVIAR: Research Results

This page links you to the key CAVIAR research results:

• Target motion detection algorithm comparison
• Gabor based interest points
• Ridge lines as novel image features
• Accurate (98%) short-term activity recognition
• HSMM-based behaviour recognition algorithm
• Methods for prioritising target detection and recognition
• Integrated behaviour recognition system
• Large ground truth for video sequence behaviour recognition research

We have two MPEG videos of some of the project results:

• The heads, hands, feet and bodies for one set of ground truth (7 Mb MPEG AVI)
• The INRIA tracker and UEDIN track connector over 200 frames of the Browse 2 sequence. (1 Mb MPEG AVI). Note there are only a few false detections that persist only a few frames.

Also possibly of interest are the PDF files of the final project review presentation:

• Overview of achievements, results on key Scientific Questions, Workpackages 2, 3, 6, 7 and management information
• Mostly workpackage 1
• Mostly workpackages 4 & 5

Back to CAVIAR home page.