基于机器视觉的动态多目标识别  

Dynamic Multi-Target Recognition Based on Machine Vision

作　　者：薛梦霞[1] 刘士荣[1] 王坚[1] 

机构地区：[1]杭州电子科技大学自动化学院,杭州310018

出　　处：《上海交通大学学报》2017年第6期727-733,共7页Journal of Shanghai Jiaotong University

基　　金：浙江省自然科学基金资助项目(LY17F030022);国家自然科学基金资助项目(61175093;61503108)

摘　　要：提出了一种基于机器视觉的实时动态多目标识别的方法.该方法首先根据前后帧之间像素的变化,分割出运动目标和样本图像,然后使用Gabor滤波器提取图像的特征,得到特征向量.最后使用Fisher判别准则分类识别,将得到的分类识别结果自动标注在输出图像中,并且将其连续输出,便能获得已经识别完成的输出视频.实验结果表明,在多个动态目标的情况下,综合运用Gabor特征与帧间差分法的动态目标识别方法能准确检测到动态目标区域,并能准确分类、识别和标注.This paper proposes a real-time dynamic multi-target recognition method based on machine vi- sion. The method first segments the moving targets according to the pixel change between the front and rear frames, then extracts the features of the segmented moving targets and sample images using Gabor fil- ter to obtain the eigenvector, and finally uses Fisher discriminant criterion for classification and recognition so as to automatically lable the classification and recognition results onto the output images. The images labeled with the recognition results are to be outputted continuously, and hence the output videos which have been recognized can be obtained. The experimental results show that in the circumstance of multiple targets, the proposed dynamic target recognition method integrating Gabor feature and interframe differ- ence can accurately detect the dynamic target area, and realize classification, recognition and labeling.

关 键 词：机器视觉 FISHER判别准则 动态多目标识别 目标分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]