Attitude Measurement Of Excavator Working Device Based On Multipoint Markers

Excavator is powerful and plays an important role in engineering construction.Generally,its working environment is very complex,which is often accompanied by noise and dust,and some occasions even have the danger of collapse and radiation.Autonomous intelligent operation of excavator has become the development trend in the future.The attitude information of excavator working device is an important parameter of smart control system.It is vital to achieve the attitude information in real time and accurately for autonomous intelligent operation.In the aspect of excavator working device attitude measurement,most of them are realized by contact sensors such as potentiometer and joint angle encoder.In recent years,some researchers have proposed a new attitude measurement method based on vision measurement technology to solve the problem that contact sensors are easy to crash and damage in attitude measurement.The artificial target with corner feature or other geometric features is pasted on the working device.The camera captures the image and processes it.After extracting the feature information of the artificial target,the attitude of the working device is calculated and the attitude measurement is completed.To some extent,there are some problems in this method.The first problem is that the artificial target is easy to be contaminated by dust,soil and other adhesion in the environment,so that the artificial features are blocked,and finally the attitude measurement of the working device fails.The second problem is that in vision plane measurement,some target features on the image will deform with the increase of camera imaging angle,which affects the detection and location of features.The third problem is that the excavator is over integrated with the working environment,and the excavator itself has edges and corners,so it is easy to extract more invalid feature information during feature extraction,and the time-consuming of feature information screening is also easy to lead to measurement failure.In view of the above problems in vision measurement,this paper proposes a vision based multi-point identification excavator working device attitude measurement system,which is equipped with a working device attitude measurement system for an excavator model without any sensors.The deep neural network is introduced into the system to effectively solve the problem of key point identification detection and location.An RGB camera and YOLOv3 deep learning algorithm are used to realize the attitude measurement of the working device without any sensors.The key point identification is designed and arranged for the working device.And it is different from the background environment and it is easy to identify and locate.Capturing and marking a large number of images of the working device to establish a data set.Based on the data set training,a visual model is obtained to realize the detection of the key point identification.According to the constraint relationship between the key points and the principle of camera imaging,the three-dimensional information of the key points is obtained,and the attitude angle of the working device is calculated.The specific research work is as follows:(1)The attitude measurement system of excavator working device is built.To design the attitude measurement scheme of working device,including the composition and layout of hardware structure;eight types of key point identification are designed and arranged,and each identification is equipped with redundant identification to ensure that the detection is not missed;to design and arrange the image data acquisition system,collect the image and make the VOC training set with sample size of 1300;based on Zhang Zhengyou’s chessboard calibration method,the calibration scheme of camera parameters is established;The contact sensor gyroscope is arranged to obtain the actual attitude angle of the working device,which can be used as the basis for evaluating the performance of the model.(2)Building the training platform and environment of two-dimensional key identification detection module.according to the data set collected and produced in the early stage,and using the YOLOv3 target detection algorithm based on deep neural network to train and obtain the key point identification detection module.(3)Aiming at the phenomenon of artificial target pollution and damage,two groups of experiments with different state identification are designed.Static measurement and dynamic measurement are carried out for each group of experiments,and the measurement results of attitude angle of working device under corresponding state are recorded and analyzed.Aiming at the phenomenon that the working device attitude angle measurement is wrong due to the deformation of the mark on the image when the camera imaging,the experiment is carried out.The research shows that the attitude measurement system of excavator working device based on multi-point identification can effectively improve some problems existing in the attitude measurement of working device based on vision;Generally,the measurement deviation range of the measurement system to the working device is [-2,2],and the average processing time for each frame image is about 108 ms,which meets the requirements of real-time measurement of the working device.