肖春明



摘 要: 由于动态摄像机行为造成了背景模糊等问题,结合角点矩特征的全局运动估计算法,将最大类间方差法引入RANSAC估计算法,对常用的RANSAC算法进行改进。在目标检测过程中,利用基于差分相乘原理的运动目标检测算法实现了运动目标的检测定位。针对获取到目标位置的后续帧序列的目标检测,提出了基于改进的奇异值分解的角点匹配的运动目标检测算法。实验表明,提出的算法对于背景发生变化的场景,能达到较好的背景运动补偿效果,在随后测量的动态背景状态下,可以准确检测后续帧序列的运动目标,并且具有良好的鲁棒性。
关键词: 目标检测; 全局运动估计; 角点; 奇异值分解
中图分类号: TN911.73?34 文献标识码: A 文章编号: 1004?373X(2016)07?0066?05
Abstract: To resolve the indistinct background problem caused by dynamic camera behavior, and in combination with the global motion estimation algorithm of angular point distance feature, the maximum interclass variance method is introduced into the RANSAC estimation algorithm to improve the common RANSAC algorithm. The moving object detection algorithm based on difference multiply principle is used to detect and locate the moving object in object detection process. Aiming at the object detection of follow?up frame sequence of the obtained object location, the moving target detection algorithm based on angular point marching of improved singular value decomposition is proposed. The experimental results demonstrate that the proposed algorithm has good background motion compensation effect in dynamic background, and can accurately detect the moving target of the follow?up frame sequence in the subsequent measured dynamic background measurement. The algorithm has good robustness.
Keywords: target detection; global motion estimation; angular point; singular value decomposition
0 引 言
运动目标检测是计算机视觉最主要的研究方向和难点之一[1?2],主要涉及到图像理解、模式识别、人工智能、自动控制和机器视觉等有关专业学科和领域的知识,并且已经被广泛地应用到很多场合 [3]。运动目标检测为后续的目标跟踪和目标识别奠定了良好的基础。在现实生活中,大量有意义的视觉信息包含在运动之中,人类可以凭借自身强大的视觉系统获取80%的信息[4],通过大脑的分析判断出运动目标,而对于计算机视觉而言,从获取视频到分析出运动的物体成为关键问题。
基于视频的目标检测系统具有结构简单、隐蔽性好、抗干扰能力强的优点,而且视频检测系统采用价格低廉的摄像机和光学成像结构,具有更高的性价比。……