改进Faster R-CNN的快件搬运机器人视觉定位策略

A Visual Navigation and Positioning Strategy for the Express CarryingRobot Based on the Improved Faster R-CNN

针对快件搬运机器人视觉导航定位精确度低的缺点,在其预设轨道上设计了一种含有4种不同颜色小矩形的导航图案,再利用改进的Faster R-CNN检测导航图案里的目标区域。然后,通过Hams角点检测算法提取目标区域的参考角点继而使用PnP算法计算出该机器人坐标和偏转角。改进的Faster RCNN是在Faster RCNN的模型基础上增加卷积层,之后在多层卷积层的feature map上生成候选框且用两个3×3卷积核分别进行卷积运算,直接进行分类和回归。最后,设计的仿真测试结果表明:两种算法检测图案中目标类别的正确率均为100%,但改进后算法的精确度提高了3%。

Aiming at the low accuracy of visual navigation positioning of the current express transportation robot,a navigation pattern with four small rectangles of different colors is designed on its preset track,and then the target area in the navigation pattern is detected by using the improved Fast R-CNN.After then,the reference corner of the target area is extracted by Harris corner detection algorithm,and the robot coordinates and deflection angle are calculated by PnP algorithm later.The improved fast RCNN is added convolution layers on basic of Faster RCNN model.Next,candidate boxes in feature maps of the multiple convolution layers are generated,and two 3×3 convolution kernels are used to do the convolution respectively for direct classification and regression.Finally,the simulation results show that the accuracy of the two algorithms is both 100%,but the accuracy of the improved algorithm is improved by 3%.