A new pedestrian detection method based on AGACS ensemble classifiers algorithm

作者：Zhang, Yang^*; Wu, Shaoxiong; Zheng, Zhaopeng

来源：Journal of Computational Information Systems, 2013, 9(23): 9269-9276.

DOI：10.12733/jcis7297

• Component classifiers
• Cost-sensitive svm
• Effectiveness and efficiencies
• Ensemble classifiers
• Kernel parameter
• Pedestrian detection
• Real time performance
• State-of-the-art algorithms

This paper presents a new algorithm named Asymmetric Gentle AdaBoost with Cost-sensitive SVM (AGACS) for the task of real-time pedestrian detection based on images and videos in machine vision field. In order to solve the problem of class-imbalanced in pedestrian detection, the proposed algorithm AGACS tries to employ Cost-sensitive SVM (CS-SVM) as weak component classifier in Asymmetric Gentle AdaBoost, which is based on the idea of assigning different weights to the errors of the two classes when the numbers of data samples from each class are imbalanced. In addition, we dynamic adjust the values of the kernel parameter in CS-SVM and get a series of component classifiers different with each other, which could improve diversity among classifiers. Finally, we measure the diversity of component classifiers, and discard poor and similar classifiers. Through above mechanism, the optimization diverse AGACS ensemble classifiers can be generated. The experiment carried out on videos from INRIA, MIT and Daimler datasets, result indicates that the effectiveness and efficiency of the proposed algorithm, which can achieve higher real-time performance and accuracy than other three state of the art algorithms.

• 出版日期2013-12-1
• 单位华南理工大学