基于显著性检测的自适应机器人拣选系统

Adaptive robot picking system based on saliency detection

在分析了现有拣选系统需要事先进行样本特征提取的情况下,为适应复杂多变的拣选环境,提出了基于显著性检测的自适应目标拣选算法。该方法通过前景目标的相互对比,识别出最具显著性特征的物体作为拣选对象,避免了预先学习的过程,并能用分析结果不断修正识别特征,提高了系统的工作效率和自动化程度。设计了适用于工业机器人的拾取控制系统,涉及网络通信、总线管理和运动控制等多方面。实验结果表明了系统的准确性与稳定性。

After the existing picking systems is analyzed, which need sample feature extraction before work, an adaptive target picking algorithm based on saliency detection is proposed to meet complex environments. The algorithm identifies the most salient object as the chosen one by mutual comparisons of the foreground objects. The pre-learning process is avoided and the results are used to revise characteristics for identification constantly. Thus, the method improves efficiency and automation of the system. A picking control system for industrial robot, involving network communication, fieldbus, motion control and many other aspects. Validation work is based on R8405 industrial robot developed by our CNC center. Experimental results demonstrate accuracy and stability of the proposed system.