RGBD SLAM Module

This module provides a comprehensive solution for RGB-D SLAM (Simultaneous Localization and Mapping) using Python. The code integrates several key libraries, such as OpenCV, Open3D, and SciPy, to process RGB-D images, perform visual odometry, and visualize results.

Features

• Loading Camera Intrinsics: Load camera intrinsics from a file.
• Combining RGB and Depth Images: Combine RGB and depth images for processing.
• Visualizing RGB-D Images: Visualize RGB-D images using Matplotlib and Open3D.
• Performing RGB-D Odometry: Perform visual odometry using Open3D.
• Sequence Processing: Process sequences of RGB-D images.
• Hybrid Visual Odometry: Combine color and depth information for odometry using feature tracking and 3D points.

Installation

1. Clone this repository:

   git clone https://github.com/hugohadfield/rgbdslam.git
   cd rgbdslam
   

2. Install with pip:

   pip install .
   

Dependencies

Ensure you have the following Python libraries installed:

• numpy
• Pillow
• click
• matplotlib
• tqdm
• pandas
• open3d
• opencv-python
• scipy

You can install these dependencies using pip:

pip install numpy Pillow click matplotlib tqdm pandas open3d opencv-python scipy

Usage

CLI Commands

The module provides a command-line interface (CLI) to process images.

Single Image Pair

Process a single pair of RGB and depth images:

python rgbdslam.py single-image-pair <rgb_path_a> <depth_path_a> <rgb_path_b> <depth_path_b>

Directory of Images

Process a directory of RGB and depth images:

python rgbdslam.py run-on-directory <rgb_path> <depth_path> <output_csv>

Example

To run the SLAM algorithm on a directory of images:

python rgbdslam.py run-on-directory rgb/ depth/ output.csv

This command processes all RGB and depth images in the rgb/ and depth/ directories and saves the results in output.csv.

Functions

load_intrinsics(intrinsics_path: str) -> np.ndarray

Load camera intrinsics from a file.

combine_rgb_depth(rgb_path: str, depth_path: str) -> np.ndarray

Combine RGB and depth images into a single image for processing.

visualise_rgbd_matplotlib(rgbd_image: np.ndarray)

Visualize an RGBD image using Matplotlib.

visualise_rgbd_o3d(rgbd_image: np.ndarray)

Visualize an RGBD image using Open3D.

extract_translation_from_4x4_matrix(matrix: np.ndarray) -> np.ndarray

Extract translation information from a 4x4 transformation matrix.

o3d_rgbd_vo(rgdb_a: np.ndarray, rgdb_b: np.ndarray, intrinsic_matrix: np.ndarray)

Perform RGBD Odometry using Open3D.

o3d_rgbd_sequence(rgbd_image_list: List[np.ndarray], intrinsic_matrix: np.ndarray, use_tqdm: bool = True)

Process a sequence of RGB-D images.

odom_extract_scipy_minimize(points_3d, points_2d, intrinsic_matrix)

Extract yaw and distance using SciPy's minimize function.

odom_extract_opencv_pnpransac(points_3d, points_2d, intrinsic_matrix)

Extract yaw and distance using OpenCV's solvePnPRansac.

hybrid_rgbd_odometry(rgbd_image_list: List[np.ndarray], intrinsic_matrix: np.ndarray, use_tqdm: bool = True, backward: bool = False)

Perform hybrid visual odometry using both color and depth information.

Contributing

Contributions are welcome! Please fork the repository and submit a pull request.

License

This project is licensed under the MIT License.

Contact

For questions or suggestions, please get in contact with Hugo Hadfield, contact details can be found on his website.