Abstract:
This paper presents an improved Gaussian mixture background model to deal with dynamic surveillance scenes. Different threshold values are utilized in the updating and detection process to reduce the misclassification rate for single mode background. In the detection process, it fuses statistical difference method and time domain finite difference method to decrease the misclassification rate for multimodal background. Experimental results show that improved model can effectively solve the mistakes of complex dynamic background, and it has good detection performance.
Key words:
object detection,
Gaussian mixture background model,
multimodal background,
parameter estimation,
data fusion
摘要: 提出一种应用于运动目标检测的改进混合高斯背景模型。在背景模型更新过程中,通过调整阈值,降低单模态背景的误检率。在运动目标检测时,融合统计差分法和时域差分法,降低多模态背景像素的误检率。实验结果表明,改进模型能有效解决由复杂动态背景引起的误检问题,具有较好的检测性能。
关键词:
目标检测,
混合高斯背景模型,
多模态背景,
参数估计,
数据融合
CLC Number:
HE Liang-Meng, QIN Rong-Hua, GONG Sai-Liang, WANG Ying-Guan. Improved Gaussian Mixture Background Model in Dynamic Scene[J]. Computer Engineering, 2012, 38(08): 10-12.
何亮明, 覃荣华, 巩思亮, 王营冠. 动态场景中的改进混合高斯背景模型[J]. 计算机工程, 2012, 38(08): 10-12.