Two-stage quick grabbing detection method based on target detection characteristics

The invention discloses a two-stage quick grabbing detection method based on target detection characteristics. The method comprises the steps of 1, obtaining a network training data set; 2, constructing a target detection convolutional neural network, a first-stage convolutional neural network structure and a second-stage convolutional neural network structure, and training three convolutional neural networks; 3, adopting three trained convolutional neural networks to obtain target types and numbers in the scene, candidate grabbing boxes of the targets and score values of the grabbing boxes; 4, obtaining an optimal grabbing box through an optimal grabbing box selection algorithm; and 5, determining the position and posture of the holder through the grabbing frame. According to the invention, the target detection method and the robot grabbing detection method are combined, and the grabbing detection accuracy and real-time performance are both considered. 本发明公开了一种基于目标检测特征的两阶段快速抓取检测方法。步骤1：获取网络训练数据集；步骤2：构建目标检测卷积神