基于Mask R-CNN视觉感知的叉车操纵器档位识别

A Vehicle Multi-manipulator Gear Recognition Method Based on Mask R-CNN Visual Perception

车辆档位信息间接反映车辆行驶状态,诸如考试、培训、安全驾驶等均需对其档位判断记录或溯源。本文以叉车为例,提出一种基于Mask R-CNN视觉感知技术的移动式、旋转式叉车操纵器档位识别方法,首先提出了基于Mask R-CNN视觉传感的叉车档位识别方法的整体架构,接着着重研究了叉车档位识别中移动式和旋转式操纵器档位识别的识别技术、叉车操纵器特征分割优化技术等,最后在一台合力K30柴油叉车上搭建试验平台进行测试。试验结果表明,该技术对叉车操纵器识别率达100%、档位判断正确率99%以上,具有无需机械改装、过程记录可溯源、准确性高、实时性较好的特点,并可拓展用于其它车辆视觉检测中。

Vehicle gear information indirectly reflects its driving status,such as examinations,training,safe driving,etc.all need to record and trace its gear.This article takes a forklift as an example to propose a mobile and rotary forklift manipulator gear recognition method based on Mask R-CNN visual perception technology,focusing on the research of the mathematical model between the mobile manipulator gear state and the calibrated gear during driving,the rotary manipulator state judgment method which can calculate the rotation angle increment of vector between the handle and the handwheel centroid in a single moment,and improve the image labeling method of forklift manipulators,this research has advantages of no mechanical modification,traceable process records,high accuracy and good real-time performance.The test results show that the research method in this paper can realize gear judgment with more than99%accuracy,and can also be applied in other vehicle visual inspection methods.