复杂场景中运动目标的检测

摘要

Because of the complexity of scene detection,traditional moving objects’extraction often adopts the methods of adaptive background update and adaptive threshold segmentation for removing noise interference and improving detection accuracy.In view of this situation,we propose to update the background through the improved background mask algorithm on the basis of background subtraction method,and to partition the detection and non-detection areas by using the complexity of scene and the feature of the system that the moving objects appear at the edge of the obstacles,as well as to make adaptive threshold segmentation on the images of whole video sequences with Pseudo threshold image method.Experimental result shows that,this method can effectively remove the noises caused by illumination changes and caused by the position deviation between the background and the current detecting frame owing to camera shake,and avoids the problem that the moving object itself appears empty when threshold is being segmented.It lays a foundation for the recognition and tracking of subsequent moving objects.%由于检测场景的复杂性，传统的运动目标的提取常常采用自适应背景更新及自适应阈值分割方法，以去除噪声干扰，提高检测准确性。针对这种情况，提出在背景减除法的基础上通过改进的背景掩膜算法进行背景更新，利用场景的复杂性及系统中运动目标出现在障碍物边缘的特点，进行检测区域及非检测区域的划分，并采用阈值伪图的方法对整个视频序列图像进行自适应阈值分割。实验结果表明，该方法能够有效去除由于光线变化产生的噪声，以及由于相机抖动引起的背景与当前检测帧之间由位置偏差而产生的噪声，并避免了阈值分割时运动目标本身会出现空洞的问题，为后续运动目标的识别与跟踪奠定基础。