多模版匹配及加权融合的行人检测方法

摘要

提出多模板匹配及加权融合的检测算法.以线性支持向量机作为分类器,采用分类器级联的训练方式,针对易错样本采用互补特征选择策略(CFSS)选择不同特征训练模板,选取红绿蓝和LUV通道下的HOG特征和细胞结构的局部二值模式(LBP)特征.利用多个模板分别进行匹配,采用线性加权的方式融合不同模板的检测结果,利用INRIA行人数据库对算法进行测试,获得更高的检测准确率.%An efficient pedestrian detection method based on multi-model matching and weighted fusion is presented.Taking linear SVMs as classifiers and classifier cascade,we use the complementary feature selection strategy to select different feature training models for error prone samples,and finally select HOG features of RGB and LUV color channels and cell-structured local binary pattern (LBP) features.The trained models are matched independently and the repeated detections from different models are combined by weighed fusion to obtain the final test results.Simulation results show that this method has more competitive performance on the INRIA pedestrian dataset.