多点标识下的挖掘机工作装置姿态测量

Attitude Measurement of Excavator Working Device Based on Multipoint Markers

针对传统视觉测量挖掘机工作装置姿态中人工靶标易被泥土污染遮挡和受背景环境影响导致姿态测量失败的问题,提出基于多点标识的挖掘机工作装置姿态测量方法,利用一台RGB摄像机和YOLOv3深度学习算法实现对无任何传感器挖掘机工作装置的姿态测量。设置区别于背景环境、易于识别和定位的关键点标识,捕获大量工作装置图像并标注建立数据集,以此数据集训练获得视觉模型实现关键点标识的检测;根据关键点间的约束关系和相机成像原理还原关键点标识的三维信息,并计算相应杆件的姿态角。研究结果表明:提出的姿态测量系统能够解决姿态测量中存在的问题,且对挖掘机工作装置姿态测量瞬时偏差在±2°范围内,平均测量偏差在±1°范围内,满足视觉伺服中姿态角的反馈;同时,测量系统对每帧图像平均处理时间为108.26ms,满足实时测量条件。

Aiming at the problem that target in attitude of excavator working device is easily obscured by soil contamination and the attitude measurement fails due to background environment impact,a multi-point mark based attitude measurement method for excavator working device is proposed,which uses an RGB camera and YOLOv3 deep learning algorithm to realize attitude measurement of excavator working device without any sensors.We have set the key point markings which are different from the background environment.Then,capture a large amount of working device image.Label the key point markings in the captured images and form the training data sets.Further,obtain the key point marking recognition model by applying YOLOv3 learning algorithm on the training data.In addition,the three-dimensional information of key point markings can be restored according to the constraint relationship between key points and camera imaging principle.Therefore,we could compute the attitude angle of corresponding bar.The results show that:this attitude measurement system can solve the problems in attitude measurement,and the instantaneous deviation of the excavator working device attitude measurement is within±2°and the average deviation is within±1°to meet the attitude angle feedback in visual servo;at the same time,the average processing time of each frame image of the measurement system is 108.26ms,which meets the real-time measurement conditions.