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。

关键词: 智能交通, 运动对象检测, 阴影消除, LBP纹理

Abstract: 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.

Key words: intelligent traffic, moving object detection, shadow elimination, LBP texture

中图分类号: 

• TP391.41