Deep Visual Inertial Odometry with Kalman Filter
Explore the advanced integration of deep learning in visual-inertial odometry in our comprehensive GitHub repository. This project highlights the development of a lightweight convolutional neural network architecture. It emphasizes the use of exponential maps and Lie groups to refine the training of neural networks with visual and inertial data, enhancing frame-to-frame navigation accuracy without relying on traditional RNNs. Dive deep into our system's technical details and practical applications in real-world autonomous navigation scenarios.
-----------------------------------------------------------------------------------
📝In-video sources
➡️ Git of Deep V-I-O: https://github.com/ElliotHYLee/Deep_V...
➡️ Git of Deep VO: https://github.com/ChiWeiHsiao/DeepVO...
-----------------------------------------------------------------------------------
🌏 Find Me Here:
🔥Linkedin: / hongyun-elliot-lee
🔥Discord: / discord
🔥ubicoders: https://www.ubicoders.com/
🔥Blog: https://www.ubicoders.com/blogs
🔥GitHub: https://github.com/ubicoders/
-----------------------------------------------------------------------------------
⏲️Time Stamps:
0:00 intro
1:17 camera calibration - traditional visual odometry pipeline
2:13 feature matching and extraction - traditional visual odometry pipeline
3:33 3D point cloud - traditional visual odometry pipeline
5:00 computing camera rotation and translation - traditional visual odometry pipeline
6:07 optical flow
7:13 flownet - optical flow CNN
8:08 camera ego-motion estimation with deep learning
10:12 extra backgrounds needed for this project
13:53 semi-supervised learning with Mahalanobis distance
19:05 demo and results
25:00 is there any better approach?
26:06 Code requirements
29:58 quick demo file
33:14 plotting result-only demo
40:26 kalman filter wrapping demo
43:10 data readers explained
45:16 models explained
#deeplearning #convolutionalneuralnetworks #computervision