Human Detection Based on a New Gradient CENTRIST Feature

Abstract:

This paper proposes a real-time and accurate human detection method base on a new Gradient CENTRIST feature descriptor. Firstly, the feature can characterizes not only local human appearance and shape but also implicitly represent the global contour. Secondly, it does not involve image pre-processing or feature vector normalization, and it only requires steps to test an image patch. Our main contribution is that a more reliable feature descriptor is found, which can get a better human detection. The experiments on the INRIA pedestrian dataset demonstrate that the detection performance is significantly improved.

• Periodical: Applied Mechanics and Materials (Volumes 543-547)
• Pages: 2716-2719
• DOI: https://doi.org/10.4028/www.scientific.net/AMM.543-547.2716
• Online since: March 2014
• Authors: Tao Li*, Tao Xiang, Ren Jie Huang, Xue Zhu Zhao
• Keywords: Gradient CENTRIST Feature, Human Detection

References

[1] A. Boudissa, J.K. Tan, H. Kim and S. Ishikawa: ICIP IEEE International Conference (2012), pp.469-472.

[2] N. Dalal, B. Triggs: CVPR IEEE Conference, Vol. 1 (2005), pp.886-893.

[3] P. Dollár, C. Wojek, B. Schiele and P. Perona: CVPR IEEE Conference (2009), p.304–311.

[4] P. Viola, M.J. Jones: Computer Vision IEEE 12th International Conference (2009), p.734–741.

[5] N.S. Vu, A. Caplier: submitted to Journal of Image Processing IEEE Transactions on (2012).

[6] B. Wu, R. Nevatia: submitted to Journal of Computer Vision (2007).

[7] J. Wu, C. Geyer and J.M. Rehe: Robotics and Automation (ICRA), IEEE International Conference on (2011), pp.860-867.