基于CNN深度学习的机器人抓取位置检测方法

针对传统检测方法受到复杂环境和人工干预影响而导致检测精准度低的问题，提出了基于CNN深度学习的机器人抓取位置检测方法。根据CNN基本结构，研究基于CNN深度学习检测原理。按照切线斜率方向划分机器人抓取位置模板点，计算模板匹配距离，得到机器模板上匹配点到边缘坐标图像点中最近的距离。保持横纵坐标变量保持不变，观察映射图上坐标灰度值及匹配度函数分布情况。引入GA求解匹配方法，根据匹配流程，寻找最优解。分析彩色图像、深度图像的可抓取位置和不可抓取位置信息，并将其转化为符合CNN深度学习的数据格式，完成信息预处理。根据机器人抓取作业示意图，设计具体检测流程，并显示检测结果，由此完成机器人抓取位置检测。由实验结果可知，该方法检测精准度最高可达到0.988，能够应用到实际机器人抓取相关任务之中。

Detection method of robot grabbing position based on CNN deep learning

Aiming at the problem that traditional detection methods are affected by complex environments and human intervention, which leads to low detection accuracy, a CNN deep learning-based robot grabbing position detection method is proposed. According to the basic structure of CNN, the research is based on CNN deep learning detection principle. Divide the robot''s grab position template points according to the tangent slope direction, calculate the template matching distance, and get the closest distance from the matching point on the machine template to the edge coordinate image point. Keep the horizontal and vertical coordinate variables unchanged, and observe the distribution of coordinate gray values ??and matching degree functions on the map. Introduce GA to solve the matching method, and find the optimal solution according to the matching process. Analyze the captureable and non-grabbable position information of the color image and depth image, and convert it into a data format conforming to CNN deep learning to complete the information preprocessing. According to the schematic diagram of the robot grabbing operation, a specific detection process is designed, and the detection result is displayed, thereby completing the robot grabbing position detection. From the experimental results, it can be known that the detection accuracy of this method can reach 0.988, and it can be applied to the tasks related to actual robot grabbing.