基于Haar-like和LBP的多特征融合目标检测算法  

Improved Multi-feature Fusion Algorithm for Target Detection Based on Haar-like and LBP

作　　者：原晓佩 陈小锋[1] 廉明 YUAN Xiao-pei;CHEN Xiao-feng;LIAN Ming(School of Automation,Northwestern Polytechnical University,Xi’an 710129,China)

机构地区：[1]西北工业大学自动化学院,西安710129

出　　处：《计算机科学》2021年第11期219-225,共7页Computer Science

基　　金：装备预先研究项目(61404130118);陕西省重点研发计划项目(2019GY-117)。

摘　　要：针对目标检测时Haar-like特征值过多、计算时间长、无法描述目标纹理特征且识别率一般的问题,提出一种基于滑窗原点信息的阈值自调节IHL(Improved Haar-like LBP)特征提取算法。该算法首先构造了IHL特征编码方法,将Haar-like特征和局部二值LBP特征融合;然后在计算Haar-like型局部二值化特征时,使用高斯矩阵获得符合像素分布规律的自调节阈值;同时在求特征值时引入中心点像素信息,确保提取的特征值的合理性;最后使用AdaBoost训练得到级联分类器,将其载入系统,并在KITTI车辆数据集和INRIA Person行人数据集上进行实验测试。实验结果表明,该方法可在65 s内识别1102个行人目标,在114.3 s内识别1852个车辆目标,相比传统算法其可以明显加快目标识别的速度,对行人和车辆目标的识别率均可达到94%以上,其检测准确性相比其他方法也有显著提升。Aiming at the problems of detecting targets with Haar-like,such as excessive feature values,expensive computational cost,inability to describe the target texture features,and low recognition rate etc.,this paper proposes an adaptive threshold IHL(Improved Haar-like LBP)feature extraction algorithm based on the information of the origin of the sliding window.More specifically,the algorithm first constructs the IHL feature coding method fusing Haar-like features and LBP features.Then,for computing Haar-like local binary features,a Gaussian matrix is used to obtain an adaptive threshold that conforms to the pixel distribution law,and the pixel information of the central point is introduced to ensure the rationality of the extracted feature value.Finally,the cascade classifier trained by AdaBoost is built and experiments are conducted on the KITTI vehicle data set and the INRIA Person pedestrian data set.The proposed algorithm,with a recognition rate of more than 94%,is proven effective with recognizing 1102 pedestrian targets in 65 s and 1852 vehicle targets in 114.3 s,which can significantly speed up target recognition time and obviously improve the detection accuracy compared with state-of-the-art methods.

关 键 词：目标检测 多特征融合 特征提取 HAAR-LIKE特征 IHL特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]