6D Pose Estimation With 3D Computer Vision Based On Deep Learning

In recent years,3D computer vision has been widely studied and applied.As an important field in 3D computer vision,the application of 6D pose estimation also is widely used in our world,such as autonomous driving,augmented reality(AR)and robotics applications,etc.In view of the background,there are two main works in this paper:(1)research on6 D pose estimation based on deep learning algorithm;(2)research on robot localization based on 6D pose.For the first work,we propose an end-to-end neural network with Feature Selecting Mechanism(FSM)for 6D pose estimation in this paper.And there are two schemes for the Feature Selecting Mechanism(FSM).The first is artificial selection and the second is variant of Autoencoder(AE).Through experiments on the Line MOD dataset,YCB-Video dataset,and our Synthetic Image dataset,the proposed method in this paper surpasses the Pose CNN and Dense Fusion.For example,on the Line MOD dataset there is an 8.1% improvement in the mean accuracy by utilizing our FSM compared to the Pose CNN and 2% improvement compared to the Dense Fusion.On the YCB-Video dataset,there is an up to 1% improvement compared to the Dense Fusion.On our Synthetic Image dataset,there is 1.9% improvement compared to the Dense Fusion.For the second work,we plan to migrate the above proposed neural network with Feature Selecting Mechanism(FSM)for 6D pose estimation to a robotic simulation platform.In this setting,we perform the research on robot localization based on 6D pose.Then,the results of three experiments demonstrate that our method is effective and suitable for several scenarios.