2020/12/24 by Daniele De Gregorio, De Gregorio, Daniele, Riccardo Zanella +5
Engineering · Computer Science · #Robot Manipulation and Learning #Hand Gesture Recognition Systems #Human Pose and Action Recognition
paper · pdf · doi:10.48550/arxiv.2012.13210
In this paper we investigate how to effectively deploy deep learning in\npractical industrial settings, such as robotic grasping applications. When a\ndeep-learning based solution is proposed, usually lacks of any simple method to\ngenerate the training data. In the industrial field, where automation is the\nmain goal, not bridging this gap is one of the main reasons why deep learning\nis not as widespread as it is in the academic world. For this reason, in this\nwork we developed a system composed by a 3-DoF Pose Estimator based on\nConvolutional Neural Networks (CNNs) and an effective procedure to gather\nmassive amounts of training images in the field with minimal human\nintervention. By automating the labeling stage, we also obtain very robust\nsystems suitable for production-level usage. An open source implementation of\nour solution is provided, alongside with the dataset used for the experimental\nevaluation.\n