Robust stereo inertial odometry based on self-supervised feature points

doi:10.1007/s10489-022-03278-w

Robust stereo inertial odometry based on self-supervised feature points

2022-07-14

2

PRE

AI

DOI：10.1007/s10489-022-03278-w

Original

Original request for help

Save

Abstract

En 中文

In the application of intelligent mobile robots, the odometry is the key system for implementing positioning. Traditional feature extraction algorithms can not work stably in challenging environments such as low-textured areas, and when the camera moves rapidly, the visual odometry can not track features. To solve the above problems, the paper proposes a robust stereo inertial odometry based on self-supervised feature points. An improved multi-task CNN is designed to extract the feature points in the images acquired by the stereo camera. In addition, we add the Inertial Measurement Unit (IMU) to cope with the rapid motion of the camera. Finally, the fixed number of key frames and IMU errors are optimized in the sliding window by minimizing a combined error function. The experimental results show that the proposed system can run in challenging scenes and maintain real-time performance. The overall performance of the proposed system is better than that of the classical stereo inertial odometry systems, and it is still competitive with the state-of-the-art methods.

Keywords:

Multi-task feature extraction network

Convolutional neural network (CNN)

Robust stereo inertial odometry

Self-supervised feature points

Robust stereo inertial odometry based on self-supervised feature points

Abstract

Journal

Organization