摘要
为了区分相机抖动导致的图像运动与真实目标导致的图像运动,提出一种新的算法,其分析了两类运动分布,而不是亮度/彩色分布。一类运动分布是长时间范围建立起来的,其仅包含背景运动。另一类运动分布是短时间范围建立起来的,既包含背景运动也包含目标运动。最终通过对这两类运动分布的差分进行阈值分割实现运动目标检测。该算法运行速度快,且内存开销低,实验表明其在有着频繁相机抖动的室内及室外场景中表现出优异的性能。
In order to deal with the disambiguation of image motion induced by the camera jitter and image motion influenced by tile truly moving objects, a new method is proposed by analyzing two types of motion distributions instead of intensity and color distributions. One type of motion distribution is accumulated during a long period of time, and contains only the background motion. The other is accumulated during a short period of time, and includes both the background motion and the object motion. Finally moving objects can be detected by thresholding the difference between the two distributions. The proposed method is fast and requires little memory. Experimental results validate the method performs well on both indoor and outdoor video sequences with heavy camera jitter.
出处
《电视技术》
北大核心
2016年第3期126-129,共4页
Video Engineering
关键词
图像处理
背景相减
目标检测
运动分布
相机抖动
image processing
background subtraction
object detection
motion distribution
camera jitter