激光与光电子学进展, 2021, 58 (2): 0210015, 网络出版: 2021-01-05
基于候选区域定位与HOG-CLBP特征组合的行人检测 下载: 870次
Pedestrian Detection Based on Combination of Candidate Region Location and HOG-CLBP Features
图像处理 选择性搜索 行人检测 完备的局部二值模式 梯度方向直方图 困难样本 image processing selective search pedestrian detection complete local binary pattern histogram of oriented gradient hard examples
摘要
基于方向梯度直方图(HOG)特征和局部二值模式(LBP)算子的行人检测算法采用滑动窗口搜索策略存在扫描区域过大和计算复杂的问题,存在的这些问题会导致检测速度慢。鉴于此,提出一种行人检测算法。首先,采用选择性搜索算法对目标区域进行定位,并将候选区域的高宽比限制在一定范围内以筛选无效窗口。然后,为了弥补LBP算子在纹理表达上的缺陷,引入完备的局部二值模式(CLBP)算子来提高纹理特征的表达能力。接着,考虑到HOG特征和CLBP算子特征维数过高对分类器的识别能力产生影响,采用主成分分析的方法分别对HOG特征和CLBP算子进行降维,降维后再进行串联融合。最后,引入困难样本的挖掘过程训练支持向量机分离器,这可以使模型训练得更充分,进而降低误检率。在INRIA数据集上仿真结果表明,所提算法在识别率和识别速度上都有一定的提高。
Abstract
The pedestrian detection algorithm based on the histogram of orientation gradient (HOG) feature and the local binary pattern (LBP) operator adopts the sliding window search strategy. The scanning area is too large and the calculation is complex, which will cause the detection speed to be slow. In view of this, a pedestrian detection algorithm is proposed. First, a selective search algorithm is used to locate the target area, and the aspect ratio of the candidate area is limited to a certain range to filter out invalid windows. Then, in order to make up for the defects of LBP operator in texture expression, a complete local binary pattern (CLBP) operator is introduced to improve the expression ability of texture features. Then, considering that the dimensionality of the HOG feature and the CLBP operator is too high to affect the recognition ability of the classifier, the principal component analysis method is used to reduce the dimensionality of the HOG feature and the CLBP, respectively, and series fusion is conducted after dimension reduction. Finally, the mining process of hard examples is introduced to train the support vector machine classifier, which can make the model more fully trained and thus reduce the false detection rate. The simulation results on the INRIA dataset show that the proposed algorithm has a certain improvement in recognition rate and recognition speed.
尧佼, 于凤芹. 基于候选区域定位与HOG-CLBP特征组合的行人检测[J]. 激光与光电子学进展, 2021, 58(2): 0210015. Jiao Yao, Fengqin Yu. Pedestrian Detection Based on Combination of Candidate Region Location and HOG-CLBP Features[J]. Laser & Optoelectronics Progress, 2021, 58(2): 0210015.