A vision-based unordered ton package structure recognition algorithm

Automatic loading and unloading is the key technology for the transfer and loading of energy storage materials. To realize the automatic loading and unloading of ton packages, it is necessary to locate and identify the ton packages first. Accurate and rapid identification and grasping of ton packages have become the focus of research. Aiming at the problems of random tonnage package posture, multi-quantity stacking, and complex grasping environment, this paper proposes a vision-based disordered tonnage package structure recognition algorithm. The RGB image is used as the grab network data sample, and the deep learning model is used to identify the center coordinate information of the ton package in the image, and the number and position information of the stacked ton packages in the scene are obtained by combining the plane search and the binocular vision depth information, to achieve efficient and accurate grabbing operation. The optimal experimental results for the detection of disordered ton packages show that the algorithm in this paper can identify all the ton package information in the image and the detection accuracy can reach 100 %, which can provide accurate information for ton package grasping.