基于局部纹理特性的运动车辆阴影消除  

Shadow Elimination for Moving Vehicle Based on Local Texture Characteristic

作　　者：祖仲林[1] 李勃[1] 陈启美[1] 

机构地区：[1]南京大学电子科学与工程系,南京210093

出　　处：《计算机工程》2009年第16期167-169,共3页Computer Engineering

基　　金：交通部科技攻关计划基金资助项目(200435333204);江苏省交通科学研究基金资助项目(05x008)

摘　　要：为解决传统阴影检测算法可靠性和实时性难以兼顾的难题,从交通场景的实际应用出发,提出一种基于局部纹理特性的灰度域阴影消除方法。通过分析阴影的物理模型,得出局部纹理的光照不变性,利用基于比值判决的LBP纹理法来区分运动车辆和阴影,并应用亮度约束和几何启发式准则进一步改善阴影检测效果。实验结果表明,该方法的阴影检测有效率在90%以上,且能较好地满足实时要求,提高低亮度时车辆的阴影检测效果。Aiming at the problem that the reliability and real-time requirement for eliminating cast shadow of moving vehicles are hard to compatible in traditional way, a new method based on local texture is proposed. Through analysis of the physical model of moving shadows, the local texture is proved to be illumination invariant. The distribution of improved LBP texture is discussed and a significant test is performed to classify each moving pixel into foreground object or moving shadow according to this theory. Intensity constraint and geometric heuristics are imposed to further improve the performance. Experiments on different scenes suggest that effective detection rate of the proposed method is over 90 percent, and the new method is still effective for moving vehicles with lower intensity and can satisfy the requirement of real-time processing.

关 键 词：智能交通 运动对象检测 阴影消除 LBP纹理 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]