基于EM聚类算法的机器人视觉场景深度分类方法  

TA classification Method of Robot Visual Scene Depth based on EM Clustering Algorithm

作　　者：王文明[1] 谭毓安[1] 

机构地区：[1]北京理工大学计算机学院,北京100081

出　　处：《信息网络安全》2013年第6期54-60,共7页Netinfo Security

摘　　要：在没有限制条件的情况下,场景及物体识别是很困难的,并且在现有的计算方法下是不可靠的。文章认为图像结构和深度存在高度依赖关系,可以根据单目视觉中的信息来判别镜头深度,这种方法不依赖于特定物体的全景结构,也不需要将图片解析成区域或物体,而只需要图像的全局结构信息,采用EM聚类算法实现,并把该方法植入机器人一体机(RIM)远程控制方案的机器人视觉中进行验证,为移动类的机器人视觉研究提供支持,拓展了机器人的视觉能力。经过试验,达到预期效果。Without limiting conditions, scene and object recognition is difficult, and the existing calculation method is not reliable. This paper holds that the image structure and depth has a closely dependent relation, put forward a kind of shot depth classification method from the information sources of monocular vision, this method does not rely on panorama structure of specific objects, does not need to parse picture into region and object, and only needs the global image s^ucture information, using the EM clustering algorithm to realize the method ,and the method is applied in robot vision of Robot Integrated Machine （ RIM ） remote control schemeI^1 and verified, provides support for mobile robot vision research, expands the robot vision ability. After the test, the system achieved the desired effect.

关 键 词：EM聚类 机器人视觉 镜头分类 场景深度 PCA降维 

分 类 号：TP242.62[自动化与计算机技术—检测技术与自动化装置]