基于Bandelet变换与人体部位的人体检测方法

Human Detection Based on Bandelet Transform and Body Parts

本文由图像的几何流特性出发,提出了一种基于第二代Bandelet变换与部位的人体检测方法.首先,利用优化后Bandelet变换的Bandelet系数及其统计特征作为人体图像的特征,通过相关试验确立了相关的最优参数和统计特征.然后再利用AdaBoost算法训练人体及各部位分类器.最后通过计算人体各部位的似然度,联合部位结合策略进行人体检测.试验结果表明,本文提出的特征提取方法能够更好地表征人体,并能有效地改善分类器性能,相应的部位检测方法可显著提高静态图像中人体目标检测的鲁棒性.

A new method is proposed in this paper for feature extraction based on geometric flow of images and the second generation of Bandelet transformation,where Bandelet coefficients and their statistical values were extracted as the feature of human images.Afterwards the full body and body parts classifier were trained on AdaBoost algorithm.At last,likelihoods of each body parts were computed combined with Bayesian decision-based approach to perform human detection.The results of human detection experiments indicate our proposed feature extraction method＇s better capability in describing human characteristics while effectively improving the performance of classifier.Combined with body parts detection,our proposed human detection method well enhanced the robustness of human detection task in both static and moving images.