C4: Real-time pedestrian detection

This is a mirror page for those who cannot visit Google Sites.

This software package detects pedestrians in real-time -- depending on the image resolution, detection speed may vary from 20fps to 100 fps on desktop computers, even 20fps on low-power embedded CPU on-board a robot.

The CENTRIST descriptor is used to catch the contour cue of pedestrians. For technical details, please refer to the following papers:

For the CENTRIST descriptor

CENTRIST: A Visual Descriptor for Scene Categorization [pdf]
Jianxin Wu and James M. Rehg
IEEE Transactions on Pattern Analysis and Machine Intelligence, 33(8), 2011: pp. 1489-1501.

For the C4 detector

Real-Time Human Detection Using Contour Cues [pdf]
Jianxin Wu, Christopher Geyer, and James M. Rehg
In: Proc. The 2011 IEEE Int'l Conference on Robotics and Automation (ICRA 2011), Shanghai, China, May 2011, pp. 860-867.
If you use the C4 software, please cite the above paper.

You can download the C4 detector in this page (C++ source code). A README is provided in the archive, as instructions to run C4.

For bug reports, inquiries, and / or suggestions, please send to me through wujx2001 AT gmail.com.
Jianxin Wu,
Apr 22, 2012 6:21 AM