timingspace / mvoscalerecovery

Recover the scale of monocular visual odometry

MVOScaleRecovery

Recover the scale of monocular visual odometry

RUN

1. save your image name in path_to_image_list by find path/| sort >path_to_image_list
2. modify the src/param.py based on your dataset
3. run python3 src/main.py path_to_image_list

Note

this is a scale recoery for a simple monocular VO, the accuracy is degraded. Current error of KITTI 00 by KITTI benchmark is 2.17% (ave every 800m)

Current result

1. KITTI 00

2. KITTI 02

3. Initial Triangles before rance

4. depth and reconstruct

mvoscalerecovery's Issues

When will this code be available?

[Question] How to run correctly with cv2.recoverPose in thirdparty/visual_odometry.py

Hello, it's an excellent work. But I face a problem when I run visual_odometry.py. It seems that there is a problem with cv2.revocerPose function.

According to opencv document opencv, the code seems reasonable for using recoverPose. So, I try to use this function without output 'triangulatedPoints' to get Rotation and translation, and it worked. I am confused why it doesn't work with output 'triangulatedPoints'.

_, self.cur_R, self.cur_t, mask,points_3d = cv2.recoverPose(E, self.px_cur,\ self.px_ref,cameraMatrix=self.camera_matrix,distanceThresh=100)

[Question]Where to get the camera height(1.75m) parameter in the dataset?

Hi， Thanks for the excellent work, i'm curious about how to get the camera height to the ground?

Will you upload the c++ version?大佬，您会公开c++版本的吗

What does norm mean in your paper?

road norm, region norm, etc.
I googled it and found an interpretation in https://en.wikipedia.org/wiki/Norm_(mathematics)#Other_norms
But I don't think the interpretation in above link is the right one.
Please tell me , thank you!

[Question] How would I use it in orb-slam3？

Hi,Thanks for the excellent work, I am a novice in ORB-SLAM,and i am learning it. I would like to know how to apply this in ORB-SLAM3?

[Question] norm calculation

Firstly, thanks for your great work.
In your paper, I can't understand the equation for calculating the normal vector and relative height.
What is the reason of normal vector obtained using inverse of 3 points in the plane rather than using cross product, and the relative height is obtained as the inverse of the size of the norm?

I will look forward to your reply
Thank you.

When will you release this code?

Last time , you said it'll release at the end of July.

[Paper link requested]

Hi, is there any http link to the paper? I want to read it.