基于改进混合高斯建模和短时稳定度的遗留物检测算法

An abandoned object detection algorithm based on improved GMM and short-term stability measure

  • 摘要: 传统遗留物检测算法存在算法过于复杂和环境适应性差的局限。本文将改进的混合高斯建模方法应用于遗留物检测,利用背景匹配失败时生成的前景模型进行前景匹配并引入短时稳定度指标,在深入挖掘前景模型中包含的遗留物信息和像素点级目标状态信息的基础上对遗留物进行综合判断。文中详细分析了传统方法的性能局限并阐述了新方法中前景模型和短时稳定度的作用原理同时给出了具体的算法流程。多场景下的实验分析表明,增加对前景模型的考察使算法在保留传统方法优点的同时具备了良好的遗留物检测能力,而短时稳定度的引入则能够进一步降低传统方法中前景模型向背景模型转换的风险。对比实验结果中本文方法在表现出良好环境适应性的同时误检团块数明显低于其他方法,算法在复杂背景条件下达到了良好的检测性能。

     

    Abstract: Traditional abandoned object detection algorithms encounter the problems of high computational complexity and poor environmental adaptability. This paper proposes an abandoned object detection algorithm based on improved Gaussian Mixture Modeling (GMM). The matching result of foreground model is considered and the short-term stability measure is employed to make a compound judgment. The information of abandoned objects contained in foreground models and the pixel-level objects status information is deeply explored to reduce the risk that foregrounds model changes into background model. The working principles of foreground model and short-term stability measure are analyzed in detail and the concrete algorithm flow is given. The proposed algorithm keeps the advantages of traditional GMM, meanwhile the foreground model and short-time stability measure enable the proposed algorithm to detect abandoned objects precisely. The experimental results under different conditions demonstrate that the proposed algorithm has better environmental adaptability and it detects much less wrong blobs than the other algorithm. The proposed algorithm achieves a better performance under complex background condition.

     

/

返回文章
返回