改进的基于模型匹配的快速目标识别

A fast object recognition method based on improved template matching

文中建立不同类型目标的模型匹配数据库;采用最小周长多边形构造目标主体轮廓的近似多边形,以简化目标主体轮廓减少算法处理的数据量;提取具有仿射不变性的多边形顶点个数、最长线段两侧顶点个数、同底三角形面积比向量特征不变量对待识别目标进行描述,应用3个特征量在模型匹配数据库中逐一进行分层遍历搜索匹配。实验表明,基于模型匹配的目标识别算法能够快速的识别目标,提高了目标识别的实时性,同时能够判定目标所处的姿态状况。

At first, the template matching database with different kinds of object model were constructed in this paper; then constructing approximate polygon of the main object contour by using minimum perimeter polygon, so that the approximate polygon can simplify the main object contour and reduce the amount of calculation; next, describing the object to be recognized by extracting three affine invariants which were the number of vertices of polygons, the number of vertices of both sides of the longest line of polygons and triangle area ratio of the same hemline respectively, then matching the three affine invariants hierarchically by using traversal search in the template matching database. At last, The results of experiment show that the object recognition algorithm based on template matching in this paper can recognize object rapidly, improving the real-time performance of object recognition and getting the posture of the object to be recognized at the same time.