Demo on PETS 2009 dataset

These are demos of crowd counting and event detection on the PETS 2009 dataset. The video is available in Quicktime format (H.264).

Crowd Counting on PETS2009

This is a demo of the crowd counting system on the PETS2009 dataset. For each video in the test set, we trained the counting system on the remaining video in the database. On the left of each video is the scene from the View001 camera. The R0 (blue), R1 (red), and R2 (magenta) regions-of-interests (ROI) are shown, along with the estimated count, standard-deviation, and [ground-truth count]. The motion segmentation is also presented: the green segments are people moving left, and the red segments are those moving right. On the right of each video is the occupancy map of the sidewalk, which is obtained by projecting the segmentation into the 3-d scene. Finally, the graph on the right plots the estimated and ground-truth counts for each ROI over time.

S1.L1: 13-57

[mov 4.1MB]

S1.L1: 13-59

[mov 3.6MB]

S1.L2: 14-06

[mov 3.1MB]

S1.L3: 14-17

[mov 1.4MB]

Event Recognition on PETS2009

This is a demo of crowded event recognition on the PETS 2009 dataset. On the top of each video is the output of the event classifiers. Green text indicates that the event was detected, while red text indicates when the event is not detected. The number in parenthesis is the probability that the event is present (the threshold was set to 0.5). The 6 event classes are: Walking (Wlk), Running (Run), Evacuation (Evc), Dispersion (Dsp), Merging (Mrg), and Splitting (Spl).