Tutorial for working with the KITTI odometry dataset in Python with OpenCV. Includes a review of Computer Vision fundamentals.
License: GNU General Public License v3.0
Hi,
First of all, congratulations! It's probably the best explanation of Visual Odommetry in the entire internet.
In your application, you used cv2.StereoSGBM_create and included that variables:
matcher = cv2.StereoSGBM_create(numDisparities=num_disparities,
minDisparity=0,
blockSize=block_size,
P1 = 8 * 3 * sad_window ** 2,
P2 = 32 * 3 * sad_window ** 2,
mode=cv2.STEREO_SGBM_MODE_SGBM_3WAY
)
But seeing the documentation in opencv website, it seems that:
P1 and P2 should be multiplied by 8*number_of_image_channels*blockSize*blockSize and 32*number_of_image_channels*blockSize*blockSize , respectively. And in your code, the images that you've used in the examples has just one channel, which means that the operation is a bit different when compared to the recommendation.
It's just a suggestion to improve your work, and I hope you see that issue with good eyes.
Again, thank you for what you've been doing. Great job!
In the Dataset_handler class, the os.listdir lists the names of the images arbitrarily, which causes the function match_features to not give accurate results as the image 000000.png (self.first_image_left) is not being compared to 000001.png (self.second_image_left) but rather another arbitrary image.
self.left_image_files = os.listdir(self.seq_dir + 'image_0')
self.right_image_files = os.listdir(self.seq_dir + 'image_1')
self.velodyne_files = os.listdir(self.seq_dir + 'velodyne')
This can be solved by sorting the names of the images generated as shown below
self.left_image_files.sort()
self.right_image_files.sort()
self.velodyne_files.sort()
Hello! Thanks for creating this amazing repo along with the videos on Youtube for beginners of KITTI odometry datasets. Here I got a question about the projection matrix in calib.txt.
In my understanding, in the videos, you said the P0-3 matrices are the rectified projection matrix, which project 3D coordinates in the associated camera coordinate system (0-3) into the global reference frame (cam0, left grayscale camera). And you prove it by decomposing the projection matrix into the intrinsic and extrinsic matrix:
P1 = np.array(calib.loc['P1:']).reshape(3,4)
k1, r1, t1, _, _, _, _ = cv2.decomposeProjectionMatrix(P1)
t1 /= t1[3]
>>> print(k1)
[[718.856 0. 607.1928]
[ 0. 718.856 185.2157]
[ 0. 0. 1. ]]
>>> print(r1)
[[1. 0. 0.]
[0. 1. 0.]
[0. 0. 1.]]
>>> print(t1)
[[ 5.37165719e-001]
[ 1.40649083e-017]
[-4.94065646e-324]
[ 1.00000000e+000]]
Then, I tried to reverse the above operation:
Rt = np.hstack([r1, t1[:3]])
P = k1 @ Rt
>>>print(P.round(3))
[[718.856 0. 607.193 386.145]
[ 0. 718.856 185.216 0. ]
[ 0. 0. 1. -0. ]]
>>>print(P1.round(3))
[[ 718.856 0. 607.193 -386.145]
[ 0. 718.856 185.216 0. ]
[ 0. 0. 1. 0. ]]
I found that the P is actually the negative of P1, it is because the function cv2.decomposeProjectionMatrix returns the -t instead of t for the translation vector. It is strange to me, could you kindly help me with this? Thanks.
Thank you so much for your help.
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
Django
The Web framework for perfectionists with deadlines.
Laravel
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
Microsoft
Open source projects and samples from Microsoft.
Google
Google ❤️ Open Source for everyone.
Alibaba
Alibaba Open Source for everyone
D3
Data-Driven Documents codes.
Tencent
China tencent open source team.