基于稀疏表达残差的自然场景运动目标检测

本文提出一种基于稀疏表达残差的非参数化运动目标检测算法,在假设前景变化相对静态背景可以视为稀疏残差的基础上,采用视频前n帧初始化稀疏表达字典;利用字典对后续视频帧进行重构,提取每帧的重构残差;结合基于光照强度的全局阈值矩阵,将残差图像二值化,提取图像前景;利用前景区域和边缘点关系剔除ghost区域;采用增量PCA(Principal Component Analysis)算法和保守更新的思想对背景模型进行更新.在changedetection.net提供的shadow数据集上实验表明,采用全局更新和残差计算的方法,可以有效的解决由于自然场景光线变化导致的阴影变化,并且对自然场景中背景的小幅度抖动和相机抖动等问题也具有一定的抵抗能力.

Moving Target Detection in Natural Scene Based on Sparse Representation of Residuals

The paper proposes a non-parametric moving target detection algorithm based on sparse representation residuals error.In order to achieve precise motion target detection,we assume that the foreground change can be seen as sparse residuals compared with the static background.First of all,we use first n frames of the video to initialize the sparse representation dictionary.It will be applied to reconstruct the subsequent frame,extract frame residuals of every image,and then extract binary foreground images combining with the pixel-based global threshold value matrix.Furthermore,we remove ghost area on the basis of the foreground and edge regions.Finally,using the incremental PCA(Principal Component Analysis)and the idea of keep and update,we renew the above background model.A set of experiments are conducted on the shadow sets of changedetection.net using global update and residual error calculation method,and the result shows that the algorithm is an effective and efficient way to adapt to changes in the shadow of a static scene because of the changes of light.What is more,as to the small amplitude changes of the static scene and camera shake problems,it can also be a good solution.