SuperPoint-Visual Odometry (SP-VO)

Abstract

This Visual Odometry is a combination between monoVO-python and SuperPointPretrainedNetwork.

Requirements

• Python 3.5
• Numpy
• OpenCV python >= 3.4
• Pytorch >= 0.4

Dataset

KITTI odometry data set (grayscale, 22 GB)

Pretrained model for SuperPoint

SuperPointPretrainedNetwork (superpoint_v1.pth)

Usage

Exec Visual Odometry

Modify the path and Sequence_Num in main.py to your image sequences and ground truth trajectories, then run

python main.py

VO Process is visualized in 2 screen (Trajectory and Feature extraction (up:SP-VO, down:Normal-VO))

Visualizer

Modify the path in result_visualizer.py to your output ("kitti_XX.txt")

python result_visualizer.py

Visualized example is following.

References

• monoVO-python
• SuperPointPretrainedNetwork
• Monocular Visual Odometry using OpenCV and its related project report Monocular Visual Odometry | Avi Singh