自动上料机器人视觉识别系统

Visual identity system for automatic feeding robot

为了满足自动上料机器人对视觉识别与定位算法高效性、高实时性的要求,提出了预检测+精检测的两步检测法.在预检测阶段,采用Bresenham圆对已提取的安全套边界点集进行曲率分类,根据边界曲率趋势筛选出特定的边界点集并求出矩形掩膜区域.在精检测阶段,在矩形掩膜区域内生成ORB特征算子检测和BRISK描述子.采用最近邻域算法进行模板匹配,利用RANSAC算法剔除误匹配.结果表明,本算法比单纯的ORB+BRISK、BRISK等算法快5~8倍;同时继承了ORB与BRISK算法的旋转不变形和尺度不变性,提升了安全套形变时顶部的识别与定位精度.

In order to meet the high efficiency and high real-time requirement for the visual identity and location algorithm of automatic feeding robot,a two-step detection method composed of pre-detection and precise detection was proposed.In the pre-detection stage,the curvature classification for the extracted boundary points of condom were carried out with the Bresenham circle.According to the trend of boundary curvature,a specific set of boundary points was selected,and a rectangular mask region was calculated.In the precise detection stage,the ORB characteristic operator and BRISK descriptor were generated in the rectangular mask region,and the template matching was performed with the nearest neighbor algorithm.In addition,the false matching was eliminated with the RANSAC algorithm.The results show that the recognization speed of proposed algorithm is 5 to 8 times faster than that of pure ORB+BRISK and BRISK algorithms.Meanwhile,the rotation and scale invariance of ORB and BRISK algorithms is inherited by the proposed algorithm,which improves the recognization and location accuracy for the top area of condom in deformation state.