visual-odometry-sfm

About

This is the repository for the project - Camera pose tracking using visual odometry. In this project, camera pose of a camera mounted on a car was tracked using visual odometry. Features extraction is done using SIFT, then outlier rejection is done using RANSAC. Then we derive the fundamental matrix, essential matrix, and camera pose from it. This pose is then plotted to obtain the trajectory.

---------- Our Implementation ------------------------------------- Opencv Implementation -----------

Video with camera mounted on the car

System and library requirements.

• Python3
• Numpy
• cv2
• math
• glob
• matplotlib

How to Run

1. Clone this repo or extract the "nalindas_proj_5.zip" file.
   
2. Navigate to the folder "Code"
   
3. Inside the python script - DataPreprocess.py, you need to specify the appropriate original dataset path in the file_path variable. Next, you need to specify the appropriate camera model parameters path in the camera_path variable. Finally, you need to specify the path where you want the processed undistorted images to be saved in the write_path variable.
4. Inside the python script - main.py, you need to specify the processed undistorted dataset path which you had specified in DataPreprocess.py in the path variable. Next, you need to specify the appropriate camera model parameters path in the camera_path variable. Finally, you need to specify the path where you want the camera pose plot images to be saved in the save_path variable.
5. In the terminal, run the command python main.py to run our implementation. Run the command python builtin_main.py to run the opencv implementation. (You need to specify the appropriate original dataset path in the IMAGES_PATH variable. Finally, you need to specify the path where you want the camera pose plot images to be saved in the MODELS_PATH variable.)