Research On Environment Mapping And Operation Object Detection For Intelligent Excavator Based On Vision-inertial Fusio

Intelligent excavation is an important development trend to cope with dangerous and harsh environments and improve the safety and efficiency of operations.Environment mapping and job object detection are two key technologies for intelligent excavators,where environment mapping refers to a good pose estimation of the intelligent excavator while understanding detailed information about the environment around the excavator,and job object detection refers to the accurate detection of the excavation target of the excavator,enabling it to complete the excavation job efficiently.Therefore,this paper focuses on two key technologies for intelligent excavators,namely environment mapping and job object detection.The specific work is as follows:(1)For intelligent excavators with large positioning errors in the face of large amounts of repetitive texture information,this paper proposes a Cross-line Theory for line feature screening based on VINS-Fusion visual-inertial fusion algorithm and replaces the ordinary optical flow method in the algorithm with a probabilistic two-stage optical flow method,and introduces line features.Then,Plücker coordinates are used to represent the line feature and a point-line feature optimization model is proposed.After the system initialization,the visual information and IMU(Inertial Measurement Unit)information are processed by nonlinear optimization.Finally,the experimental results of this paper algorithm are compared with those of the VINS-Fusion and PL-VINS algorithms on the Eu Ro C dataset in terms of accuracy.The experiment shows that the proposed algorithm is more accurate in feature extraction and pose estimation,and a dense mapping thread is added to show the mapping effect.(2)This study utilizes the Mask RCNN object detection algorithm to accurately detect the common excavation objects,stones.Firstly,a dataset of stones is constructed by field investigations and large amounts of data collection.After that,considering the deficiencies of the Mask RCNN algorithm,the feature pyramid network is improved and a image fusion module is added.Finally,training and testing on the self-built dataset proves that the algorithm proposed in this paper is more accurate for the detection of stones.(3)In an excavation machine actual working scene,the optimized Visual-Inertial Fusion SLAM(Simultaneous Localization and Mapping)algorithm is tested.Multiple outdoor experiments all show that this algorithm not only can effectively improve the positioning accuracy,but also meet the real-time requirements.At the same time,dense mapping thread is used to complete the map construction of the excavation machine working surface.Finally,stone detection is carried out on the excavation machine working surface,verifying the feasibility of the improved algorithm in the actual scene.