Supplemental Results of Our Method for Automatic Detection of Interaction Groups

Introduction:
In this page we present some results of our proposed methods for automatic detection of multi-object interaction groups and its application to hot-spot event detection.
 
Result A: Quantitative Results on Two Datasets (Click to View Larger Images)
1500-1600
(a) Positive Recall Rate of Ground-truth Interactions
1500-1600
(b) Overload Rate of Objects per GT Interaction
1500-1600
(c) Overload Rate of Interactions per GT Interaction
 
Result B: Robustness Against Noisy Trajectories (Click to View Larger Images)
noisy-robust
(a) Results with pre-smoothing (21-frame temporal window)
noisy-robust
(b) Results with pre-smoothing (41-frame temporal window)
 
Result C: Recalling Rate of Hot-spot Events in the JAIST MVS Dataset
noisy-robust
 
Result D: Failed Cases in the JAIST MVS Dataset
noisy-robust
 
Result E: Demo Video of Our Results on A Long Video From the JAIST MVS Database
Demo video
Download High Quality Full Resolution Version of the Above Demo Video (~250MB)
 
Demo Video of Our Results on A Long Video From the APIDIS Basketball Database
Demo video
Download High Quality Full Resolution Version of the Above Demo Video (~300MB)

Maintained by chen-fan AT jaist.ac.jp. Last update 2012-10-20

 

Valid HTML 4.01 Transitional