Proceedings of the
9th International Conference of Asian Society for Precision Engineering and Nanotechnology (ASPEN2022)
15 – 18 November 2022, Singapore
doi:10.3850/978-981-18-6021-8_OR-02-0280
Vision-based Payload Volume Estimation for Automatic Loading
Singapore Institute of Manufacturing Technology, A*STAR, 2 Fusionopolis Way, Singapore 138634, Singapore
ABSTRACT
In construction industry, the number of skilled workers is decreasing due to aging and avoidance of labor-intensive work. In addition, construction accidents take a large proportion of occupational accidents, about half of which are related to construction machinery such as excavators. To solve these problems, the need for studies on automation of excavator is increasing. Excavator work is largely divided into excavation and loading. Loading is important because it is a large part of excavator work and loading a large amount of payload in a single trip is an enormous gain across the entire job. Therefore, measuring volume of payload is essential to ensure high loading rate while preventing overloading. However, it usually depends on operators’ observation based on their experience and can be challenging when the surface of payload is rough. In that sense, automatic volume estimation methods using pre-stored container model and multiple sensors have been developed, but they are not feasible to construction work. Therefore, this paper proposes payload volume estimation algorithm using a single depth camera. The proposed algorithm recognizes the truck dump body in RGB image using instance segmentation. With this information, we get the region of interest in dump body and estimate payload volume by processing point cloud. We use iterative closet point (ICP) and density-based spatial clustering of applications with noise (DBSCAN) algorithm to calculate the pose of dump body and improve the error of payload volume. Then, loading is executed according to the estimated payload. The algorithm is implemented and evaluated in RC excavator equipped with a depth camera under the boom and result of experiment show that the algorithm produces accurate estimation of payload volume.
Keywords: Loading, Volume estimation, ICP, DBSCAN
Singapore Institute of Manufacturing Technology, A*STAR, 2 Fusionopolis Way, Singapore 138634, Singapore
ABSTRACT
In construction industry, the number of skilled workers is decreasing due to aging and avoidance of labor-intensive work. In addition, construction accidents take a large proportion of occupational accidents, about half of which are related to construction machinery such as excavators. To solve these problems, the need for studies on automation of excavator is increasing. Excavator work is largely divided into excavation and loading. Loading is important because it is a large part of excavator work and loading a large amount of payload in a single trip is an enormous gain across the entire job. Therefore, measuring volume of payload is essential to ensure high loading rate while preventing overloading. However, it usually depends on operators’ observation based on their experience and can be challenging when the surface of payload is rough. In that sense, automatic volume estimation methods using pre-stored container model and multiple sensors have been developed, but they are not feasible to construction work. Therefore, this paper proposes payload volume estimation algorithm using a single depth camera. The proposed algorithm recognizes the truck dump body in RGB image using instance segmentation. With this information, we get the region of interest in dump body and estimate payload volume by processing point cloud. We use iterative closet point (ICP) and density-based spatial clustering of applications with noise (DBSCAN) algorithm to calculate the pose of dump body and improve the error of payload volume. Then, loading is executed according to the estimated payload. The algorithm is implemented and evaluated in RC excavator equipped with a depth camera under the boom and result of experiment show that the algorithm produces accurate estimation of payload volume.
Keywords: Loading, Volume estimation, ICP, DBSCAN
