Shape-Based Pedestrian Detection using a Novel Hierarchical Structure

摘要

This paper proposes a new algorithm that implements a shape-based hierarchical structure for robust pedestrian detection. Unsupervised template clustering is used to implement a binary tree-like structure. Simple and effective silhouette features are introduced for shape description and similarity analysis based on an existing Shape Context algorithm. The revised Shape Context features are combined with the Ada Boost algorithm, which is used for feature selection. The new algorithm greatly reduces the computation complexity of the original algorithm. Experimental results demonstrate the high detection accuracy and effectiveness of the proposed algorithm compared with other popular algorithms.