摘要
针对多运动目标的识别与计数,首先提出了一种改进的像素标记法,即对二值图像进行逐点扫描,将同一目标的像素赋予相同的标号,形成相应的标签图像,并初步统计出运动目标的数量;通过标记后,进一步按照运动目标的标号,提取每一个运动目标的面积、高宽比、质心特征量,并通过阈值判定,将特征量不在阈值范围内的伪目标清除,最终实现真实运动目标数量统计。实验证明该方法稳定性强、准确率较高,针对轮廓复杂的目标也有良好的效果。
Firstly a modified method of pixel marker is proposed aimming at recognition and counting of moving objectives. This algorithm will be applied to acquire object features and segment image accurately. That is, scanning binary image point - by - point, giving identical markers to the pixels of a tanget to form a corresponding labelled image so that preliminary number of moving targets can be conunted using the marker of moving target. After labelling, the area, depth - width ratio and centroid eigenvalue of every moving object can be extracted according to the markers of the objectives. And through the threshold determination, pseudo - objective whose eigenvalue is not within the threshold will be cleared and eventually realize the statistics of the number of targets. Experiment proved that the method is more stable and has higher accuracy, and is suitable for the targets having complex contours.
出处
《计算机仿真》
CSCD
北大核心
2009年第8期199-203,共5页
Computer Simulation
关键词
二值图像
像素标记
特征提取
多目标计数
Binary image
Pixel marker
Character extraction
Multi - objective counting