A Benchmark for Camera Systems that Monitor Pedestrians
ristani at cs.duke.edu
||Tuesday, April 19, 2016
||1:00pm - 2:00pm
||N311 North, Duke
To help accelerate progress in multi-target, multi-camera tracking systems, we present (i) the largest fully-annotated and calibrated data set to date with more than 1 million frames of 1080p, 60fps video taken by 8 cameras observing more than 2,800 identities over 85 minutes; (ii) a new pair of precision-recall measures of performance that treats errors of all types uniformly; and (iii) a plug-and-play, architecturally uniform reference software system for the experimental evaluation of system components. We show that (i) our data set poses realistic challenges to current trackers; (ii) our measure of performance matches ground-truth and computed identities more consistently than existing measures in the multi-camera setting; and (iii) the performance of our reference system is comparable to the state of the art.
Advisor(s): Carlo Tomasi
Committee: Pankaj Agarwal, Ronald Parr