Object recognition and 6D pose estimation using deep learning

• Main Author: Sivadas Thinagar Nanoo
• Other Authors: Teoh Eam Khwang
• Format: Final Year Project
• Language: English
• Published: 2019
• Subjects: DRNTU::Engineering::Electrical and electronic engineering
• Online Access: http://hdl.handle.net/10356/77466
• Institution: Nanyang Technological University

Object recognition and 6D pose estimation are imperative for robots to relate to the real world. However, due to occlusion, clutter and the properties of various objects in a scene, it might be challenging and tedious for a robot to recognize and estimate the 6D pose of objects. Various methods have been presented throughout the years with concern to this topic. However, many of these methods have its set back and limitations. Due to these reasons, rose the motivation to develop a robust and versatile real time system capable of accurate object recognition and 6D pose estimation with respect to the industrial standards. Over the years, with the advancement in technology, computing power have improved drastically. Algorithms, techniques and methods that were once infeasible to implement due to high computational power requirements could now be done with ease. One such implementation is none other than deep learning. It is now the current state-of-the-art technology. Since, this project is in relation with computer vision, the deep learning architecture proposed would be a convolutional neural network. Hence, the dataset used would consist of images. A deep learning framework known as PoseCNN is explored for object recognition and 6D pose estimation capabilities. In this project, a detailed literature review of PoseCNN, as well as comparisons with current approaches, will be reviewed. Finally, the results obtained using the YCB dataset would be presented.