基于Dirichlet过程的行人检测方法

PEDESTRIAN DETECTION METHOD BASED ON DIRICHLET PROCESS

使用图像特征提取技术、特征描述技术、特征码书和Dirichlet过程,自适应地生成了一组部件来表征行人。在此基础上,结合了均值漂移算法、部件直方图和支持向量机等技术提出一种新颖的行人检测方法。实验结果表明该方法结构简单,具有较高的检测率和较低的误检率,同时还可以推广至检测其他类型的物体,具有较高的实用价值。

Using the techniques of image feature extraction and feature description, as well as feature codebooks and Dirichlet process, we can adaptively obtain a group of parts to characterize pedestrians. Based on these common techniques ,we present a novel method to detect pedestrians by additionally introducing mean shift algorithm, histogram of parts and support vector machine. Experiments. show that this method has simple framework, high precision rate and low false positive rate;it can also be generalized to detect other class of objects and has fairly high practical value.