一种基于梯度规范化的目标检测算法

An Object Detection Algorithm Based on Normalize Gradients

针对传统的目标检测任务都是基于滑动窗口模型来提取特征的,由于产生过多冗余的窗口会显著增加计算量并且对后续的特征提取和分类任务造成影响,提出一种选择性搜索结合梯度规范化的特征提取算法,首先利用选择性搜索算法产生一组目标区域集,该算法能够大幅度降低搜索空间并产生较少的目标区域,然后利用梯度规范化对目标区域进行特征映射。

In view of the fact that the tradit ional object detection task is based on sliding window model to extract the features, which may produce so many redundant windows that will significantly increase the amount of calculation and affect the subsequent feature extraction and classification task In order to present an extraction algorithm which combines selective search and gradient standardization. First, selective search algorithm is used to produce a set of the object areas, which can greatly reduce the search space and produce fewer object areas. Then the object areas are mapped by exploiting the gradient feature normalize. The detection accuracy will be improved by classifying these feature vectors.