Pedestrian Detection
The goal of this project is to build pedestrian detectors with low false-positive and high detection rates, which can operate in real-time. We combined integral channel features with our ECBoost algorithm for building cascaded detectors. Below we showed effect of each feature and performance comparison with state-of-the-art on caltech pedestrian dataset.
Effect of different channel of features:
Comparison with state-of-the-art for 100 pixel pedestrian:
Comparison with state-of-the-art for 50 pixel pedestrian (Context is our result):
Video demos from Caltech pedestrian dataset:
Download	Download
Video demos from UCSD campus:
Download	Download
Download	Download
Download	Download
Related Publications:
Learning Optimal Embedded Cascades Mohammad J. Saberian and Nuno Vasconcelos.  IEEE Transactions on Pattern Analysis and Machine Intelligence   vol. 34(10), 2005-2018, October 2012 [ps] [pdf]   Boosting Classifer Cascades Mohammad J. Saberian and Nuno Vasconcelos . In Proc. Neural Information Processing Systems (NIPS), Vancouver, Canada, Dec 2010. [ps] [pdf]

Copyright @ 2009 www.svcl.ucsd.edu