DecoupledPoseNet: Cascade Decoupled Pose Learning for Unsupervised Camera Ego-Motion Estimation

Although many impressive works on learning-based camera ego-motion estimation methods have been proposed recently, most of them promote the accuracy of camera pose estimation by various sequential learning with loop closure optimization, while neglecting the improvement of PoseNet itself. In this paper, we focus on the coupling of rotation and translation in ego-motion estimation, and design a cascade decoupling structure to separately learn the rotation and translation of camera relative motion between adjacent frames. Meanwhile, a rigid-aware unsupervised learning framework with iterative pose refinement scheme is proposed for camera ego-motion estimation. It can disambiguate rigid motion and deformations in dynamic scenarios by jointly learning of optical flow, stereo disparity and camera pose. Validated with evaluation experiments on the public available datasets, our method is superior to the state-of-the-art unsupervised methods, and can achieve comparable results with the supervised ones.

I. Introduction

Camera ego-motion estimation or visual odometry (VO) has drawn great attention of the researchers due to its crucial role in many visual tasks, such as VR/AR, 3D modeling, visual simultaneous localization and mapping (vSLAM), etc [1], [2]. It takes a sequence of consecutive images (or video frames) recorded by a moving camera as input, and recovers the relative movement of camera (ego-motion), which is usually parameterized as 6-DoF camera pose (translation and rotation) [3], [4]. In the last two decades, traditional methods [5], [6], based on the multi-view geometry theories, have shown their impressive performance in well-conditioned environments. However, their performances and robustness are easily affected by the textureless regions or the low image-quality. Moreover, traditional methods are typically computationally expensive because of the complicated optimization.