基于运动轮廓标记提取的分水岭运动目标检测

针对运动目标差分相乘产生的空洞问题,提出了一种基于运动区域轮廓信息和自适应标记约束的分水岭运动目标检测新算法,来实现移动机器人平台上较大运动目标的完整检测.首先,结合相位相关法和Fourier-Mellin变换配准图像的缩放和平移量;然后,利用运动区域轮廓信息和分层投影法来提取前景和背景标记.先通过连续三帧配准图像差分相乘方法检测出运动区域轮廓,并结合形态学腐蚀、膨胀操作和投影法生成前景和背景标记模板;再将前景和背景标记模板分为若干层,通过水平投影得到每一层轮廓的边界点,并按一定方式连接得到前景和背景标记;最后,根据重构的梯度图像,用标记约束分水岭分割出完整的运动区域.实验结果表明,该算法能够准确完整地分割出规则和非规则运动目标,具有较好的实时性.

Moving object detection by marker-extracted watershed based on motion contour

Aiming at the problem of empty caused by using difference multiplication to a moving object, a new algorithm is introduced based on motion region contour and adaptive marker-constrained watershed to realize the integrity of bigger moving object detection on a mobile robot platform. Firstly, scaling and translating of background caused by the mobile robot are matched using the phase correlation and Fourier-Mellin transform algorithm. Secondly, the foreground and background markers are extracted based on motion region contour and layered projection method. The motion region contour is detected by multiplying the two difference images of three frames. And then, through the combination of the open, close operation and projection method, the foreground and background marker templates are generated. These marker templates are divided into several layers, and the foreground and background markers are extracted via connecting the boundary points of each layer in a certain way, these boundary points are gained by horizontal projection. Finally, based on the reconstructed gradient image, the motion region is completely segmented using the marker-constrained watershed algorithm. The experimental results demonstrate the effectiveness of the algorithm for segment of ordered and dis-ordered moving objects and its real-time performance.