多级分块的交通视频智能识别背景建模方法

发布时间：2018-04-24 23:02

本文选题：背景建模 + 帧差法　；参考：《哈尔滨工业大学学报》2017年09期

【摘要】：为改善视频监控中的背景建模方法对于前景目标物较多或者光线变化的复杂环境效果不太理想的缺陷,提出一种多级分块背景建模方法.该方法以间隔N帧帧差法为基础,采用多级分块,并结合对称二值模式(center-symmetric local binary pattern,CSLBP)和码本(codebook,CB)等算法建立背景模型.通过模型得出背景较为清晰和完整,为下一步进行前景目标的准确识别提供良好基础.采用设计实验检验该方法的有效性,将其与局部二值模式(local binary pattern,LBP)、CSLBP、CB以及经典的混合高斯背景建模(mixture of Gaussian,MOG)等算法进行对比分析,得出采用此方法提取的前景目标物更加完整,边界更加清晰,且无明显分块图形出现.采用评分的方法对几种方法进行综合评分,该方法评分较高.在对前景目标物的提取方法中,该方法效果较好.
[Abstract]:In order to improve the defects of background modeling method in video surveillance, which is not ideal for the complex environment with more foreground objects or light changes, a multi-level segmented background modeling method is proposed. This method is based on the spaced N-frame difference method, uses multi-level block, and combines symmetric binary local binary pattern local (CSLBPN) and codebook (CBB) algorithms to establish the background model. The background is clear and complete through the model, which provides a good basis for the accurate recognition of the foreground target in the next step. The effectiveness of this method is verified by design experiments. The algorithm is compared with the local binary pattern LBPU / CSLBPU CB and the classical hybrid Gao Si background modeling mixture of Gaussian MOG. It is concluded that the foreground target extracted by this method is more complete. The boundary is clearer and there is no obvious block pattern. Using the method of scoring several methods for comprehensive evaluation, this method has a higher score. In the extraction of foreground object, the method is effective.
【作者单位】：西南交通大学交通运输与物流学院;综合交通运输智能化国家与地方联合工程实验室;
【基金】：国家自然科学基金(71402149,51578465) 西南交通大学博士创新基金(YH1000212471404)
【分类号】：TN948.6;TP391.41

【相似文献】