摘要
针对物体检测容易受外形、视觉等可变性影响的问题,提出一种利用支持向量回归(SVR)模型的区域池化检测方法,即自适应区域池化方法,该方法适用于分割区域,能自动发现不同的实例和图像块。生成区域方案,每种样本的方案都由不同颜色的边界框表示;利用区域池化法提取特征,解析区域结构;分类数据采用非极大值抑制法得到检测结果。实验结果验证了提出方法的有效性,与其它同类方法相比,该方法对物体检测的性能明显提升,其中平均召回率达到了90.8%,加入CNN特征,性能提升幅度更大。
Concerning the variability of the shape and vision of objects,a method for object detection using the regional pooling of support vector regression（SVR）model was proposed,namely adaptive regional pooling method.The method was suitable for the segmentation region,and it can automatically find different instances and image blocks.The region scheme was generated,and each sample was represented by the boundary frame of different colors.The feature was extracted using the method of regional pooling,and the structure of the analytic area was analyzed.The non-maximal suppression method was adopted for the classification data to get the test results.The effectiveness of the proposed method was verified by the experimental results.Compared with other similar methods,the performance detection of objects is significantly improved.The average recall rate of the proposed method reaches 90.8%.By adding CNN features,the performance improvement is more significant.
出处
《计算机工程与设计》
北大核心
2017年第10期2804-2808,共5页
Computer Engineering and Design
基金
国家自然科学基金项目(60974016)
江苏省自然科学基金项目(BK20131097)
江苏高校品牌专业建设工程基金项目(PPZY2015C239)
关键词
区域池化
自适应
支持向量回归
非极大值抑制法
分类
regional pooling
adaptive
support vector regression
non-maximal suppression method
classification
