基于自适应纹理特征和码本模型的分层运动目标检测算法研究  

Research on Layered Moving Target Detection Based on Adaptive Texture Features and Codebook Model

作　　者：马健博 姜肇国 张宁[1] MA Jian-bo;JIANG Zhao-guo;ZHANG Ning(School of Optoelectronic Engineering,Changchun University of Science and Technology,Changchun 130022)

机构地区：[1]长春理工大学光电工程学院,长春130022

出　　处：《长春理工大学学报（自然科学版）》2020年第6期8-14,共7页Journal of Changchun University of Science and Technology(Natural Science Edition)

基　　金：吉林省科技发展计划项目(20170204048GX)。

摘　　要：如何在复杂的环境中提高运动目标检测的准确率是视频分析领域的重要研究课题。提出一种自适应纹理特征(自适应SILTP算子)和码本模型(CodeBook)结合的分层运动目标检测的算法。首先在区域级利用自适应SILTP算子进行第一层运动目标检测,分析运动目标的像素间的空间分布相关性并具有光照变化鲁棒性,确定运动目标所在区域,然后进行精确检测,在像素级采用码本模型进行第二层运动目标检测,通过检测粒度的逐步细化,逐渐逼近运动目标,从而获得更为精确的运动目标。通过试验表明分层运动目标检测算法具有很好的准确度和鲁棒性,不仅保证了运动目标检测结果的准确和轮廓的完整,而且在光照、阴影等复杂环境中也有鲁棒的检测结果。It is an important research project in the field of video analysis that how to improve the accuracy of moving target detection in complex environment.In this paper,a layered moving target detection algorithm based on adaptive texture feature(adaptive SILTP operator)and codebook model(CodeBook)is proposed.Firstly,the adaptive SILTP operator is used to detect the first level moving object at the regional level;and the spatial distribution correlation of the moving object pixels is analyzed,which has the robustness of illumination change.Then the region of the moving object is determined;the precise detection is carried out.At the pixel level,the second level moving object is detected by the codebook model.Through the gradual refinement of the detection granularity,the motion is approached gradually in order to obtain more precise moving targets.The comparative experiments show that the hierarchical moving object detection algorithm has good accuracy and robustness,which not only ensures the accuracy of moving object detection results and the integrity of the contour,but also has robust detection results in complex environments such as light,shadow.

关 键 词：运动目标检测 自适应SILTP 码本模型 分层算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]