点击次数：

论文类型：期刊论文

发表时间：2010-10-01

发表刊物：ICIC Express Letters

收录刊物：EI、Scopus

卷号：4

期号：5 B

页面范围：1931-1936

ISSN号：1881803X

摘要：An automatic and robust approach for pedestrian detection is proposed in natural scene images, even in the presence of clutter and occlusion. An improved statistical descriptor is designed to encode more support vectors, and the dimensionality reduction method compactly optimizes it without its performance degradation. To improve the efficiency, similarity measurement kernel adopts a pyramid matching mode, and the statistic multi-models detector significantly improves the performance on challenging scenes. Comparisons in various aspects validate that the improved descriptor is more distinctive and robust, which results in a higher discriminant performance and a better efficiency. The experimental results testify the effectivity and robustness of pedestrian detection in complicated scene images. ? 2010 ICIC International.