Visual-Odometry-on-Monocular-Camera

This project implements Visual Odometry using ORB features and the KITTI dataset to estimate camera pose from image sequences.

Overview

The VisualOdometry class provides core functionality:

1. Initialization: Loads calibration data, ground truth poses, and images.
2. Feature Detection and Matching: Uses ORB to detect features and FLANN matcher for descriptor matching.
3. Pose Estimation: Computes the Essential matrix, decomposes it to find rotation and translation, and estimates camera pose.

Detailed Explanation

1. Initialization (__init__):

   • Load calibration (K and P), ground truth poses, and images.
   • Initialize ORB detector and FLANN matcher.

2. Load Calibration Data (_load_calib):

   • Reads calibration file for intrinsic matrix K and projection matrix P.

3. Load Ground Truth Poses (_load_poses):

   • Reads and formats ground truth poses as 4x4 matrices.

4. Load Images (_load_images):

   • Loads grayscale images from the specified directory.

5. Feature Matching (get_matches):

   • Detects and matches ORB keypoints between consecutive frames.
   • Filters matches using Lowe's ratio test.
   • Displays matches for visualization.

6. Pose Estimation (get_pose):

   • Estimates Essential matrix and decomposes it to retrieve rotation R and translation t.
   • Constructs a transformation matrix.

7. Decompose Essential Matrix (decomp_essential_mat):

   • Decomposes the Essential matrix into possible transformations.
   • Triangulates points to select the best transformation.

8. Main Function (main):

   • Initializes VisualOdometry with dataset.
   • Iterates over poses, matches features, and estimates pose for each frame.
   • Visualizes estimated path against ground truth.

Running the Code

1. Place the KITTI dataset in a directory (e.g., VisualSLAM/KITTI_sequence_1).
2. Ensure the directory contains calib.txt, poses.txt, and the image_l directory.
3. Run the script:

   python your_script_name.py

The system will display feature matches and plot the estimated path against the ground truth.