Diverse cost-sensitive asymmetric SVM-boost classifier with full binary tree structure for fast pedestrian detection system

作者：Yang, Zhang^*; Weiming, Liu; Rufeng, Deng; Chen, Mo

来源：Journal of Convergence Information Technology, 2011, 6(8): 117-125.

DOI：10.4156/jcit.vol6.issue8.14

• Asymmetric adaboost
• Cost-sensitive
• Diversity
• Effectiveness and efficiencies
• Gentle AdaBoost algorithm
• Pedestrian detection
• Pedestrian detection system
• Series connection structures

A high real-time and robust performance pedestrian detection method based on machine vision is proposed in this paper. Firstly, the novel features called Locally Assembled Binary Haar (LABH) are selected as the feature vector. In these features, Haar features keep only the ordinal relationship named by binary Haar feature, and assemble several neighboring binary Haar features to improve the ability of illumination invariant and discriminating power. Furthermore, a new classifier algorithm named Diverse Cost-sensitive Asymmetric SVM-Boost (DCSASB) is proposed. In this algorithm, we try to use RBFSVM (SVM with the RBF kernel) as component weak classifiers, based on dynamic regulation the values of kernel parameter σ and the penalty parameter C, a series of component classifiers different with each other can get, and most of these classifiers have moderately accurate. Then, we improve Asymmetric Gentle AdaBoost algorithm, trains diverse RBFSVM classifiers and brings in a principle of cost-sensitive at the same time. Finally, a Full Binary Tree structure is presented to build on an efficient classifier structure, which has advantages of both series connection structure and parallel connection structure, and brings in a principle of "Early-rejection" could improve system's real-time performance. The experiment carried out on videos from INRIA, MIT and Daimler datasets demonstrated that the effectiveness and efficiency of the proposed method in complex scene urban environment.

• 出版日期2011-8
• 单位华南理工大学