Grasping Detection Based on YOLOv3 Algorithm

Abstract

In order to achieve multi-object grasping detection, this paper draws on the YOLOv3 algorithm, which has good performance in the field of object detection, and designs a grasping detection model based on deep learning. This model can recognize and locate multi-object in real time, and can predict the grasping point and the grasping deflection angle at the same time, which can realize end-to-end object positioning and grasping detection. For the specific application scenario, the corresponding dataset is automatically produced through data enhancement, which greatly saves the cost and time of manual collection and labeling. In the actual scene, we identify and locate the object and estimate the grasping point and grasping deflection angle of the object at the same time. 99.5% of the objects can be detected on the self-made data set. This method can greatly improve the accuracy of grasping detection in a specific scene.