基于改进Mask R-CNN的纸箱堆垛分割与定位方法

Carton detection and localization method based on the improved Mask R-CNN

我国物流仓储行业发展迅速,利用仓储机器人实现仓库无人化是发展趋势,如何使机器人自主识别纸箱并进行定位具有重要的研究意义。为提升图像处理网络模型的性能,以完成对纸箱的分割与定位任务,对现有网络存在的不足进行理论分析,并结合仓储环境下纸箱的特点和机器人作业的要求,提出一种Mask R-CNN网络的改进方法。针对漏检误检率不理想和定位不精准等问题,分别提出了引入上下文机制和改善损失函数等改进方案,实验证明,改进后的Mask R-CNN网络在识别率与定位准确率上都有大幅提高。

With the rapid development of logistics and warehousing industry in China,it has become a trend to use the storage robots in unmanned warehouse.How to make the robots detect cartons and locate them automatically is of great significance.In order to improve the performance of the image processing network model and complete the task of carton detection and localization,the shortcomings of the network are analyzed theoretically,the cartons characteristics and the robot operation requirements in the storage environment are taken into consideration,and an improved Mask R-CNN network method is proposed.In view of the problems such as error detection and inaccurate localization,the introduction of contextual information and the improvement of loss function are proposed.The experiment shows that the Mask R-CNN network has a significant improvement in recognition rate and localization accuracy.