基于增量式特征基背景模型的运动目标检测

背景差法是目标运动检测的主流方法,关键在于背景模型自适应更新.针对传统特征基背景模型批处理方式计算量大、更新速度慢的问题,采用增量式主成分分析来建立特征基背景模型.首先计算样本图像的初始背景图像,然后采用CCFIPCA算法更新特征基背景模型,最后通过输入帧和重建帧的欧氏距离检测前景运动目标.算法以视频帧整体来建立背景模型,克服了混合高斯模型和核密度估计以孤立像素点建模的不足,提高了背景建模的鲁棒性.在SIMULINK下的仿真实验表明,算法能很好地适应高速公路交通场景动态变化,在有光线变化和阴影影响的情况下能完整、准确地提取出运动车辆轮廓.

Moving object detection using incremental eigen background model

The background subtraction is an important method to detect the moving object, and its key is background model adaptive update. To deal with the problem of compute complexity and update difficulty with traditional batch eigen background model, a background reconstruction algorithm based on candid covariance free incremental principal component analysis is introduced in this paper. Firstly, the mean background of initialize samples is obtain, and then update the eigen background model using candid covariance...