KinectFusion

Setup Environment

conda create --name kinect python=3.10
conda activate kinect
pip install -r requirements.txt

Pre-requirements

1- Download the dataset freiburg1_xyz and store it in Data folder in the parent folder of the KinectFusion/ folder. https://vision.in.tum.de/rgbd/dataset/freiburg1/rgbd_dataset_freiburg1_xyz.tgz

2- Create folder mesh_out in the base of KinectFusion

3- Your directory structure should be as follows

Project/ 
-> Data/rgbd_dataset_freiburg1_xyz/
-> KinectFusion/
    -> mesh_out/
    -> code files

4- To run simply

python main.py

Documentation

main.py

Takes in the arguments and initializes the dataset loader, VirtualSensor and Parser possible arguments are

--visualizeTSDF (Visualizes the TSDF after 700 frame)
--ground (Uses the ground truth instead of the ICP)

dataset.py

Loads our dataset by parsing the folder content and storing the paths for the images as well as storing the ground truth

virtualSensor.py

Processes the frame by loading the depth and rgb images as well as loads the current trajectory ground truth

camera_sensors.py

Stores the camera intrinsics and the image sizes

kinect_parser.py

Initializes the TSDF Volume with our camera intrinsics, responsible for the main loop of the application by calling the virtualsensor to process the next frame, then calls the Layer to generate the vertices and normal maps, then initializes the TSDF fusion first frame, and following frames it calls the ICP then integrates the TSDF. It is also responsible for exporting and viewing the TSDF

layer.py

Responsible for the surface measurements.

transforms.py

mainstreams the transformation between spaces

tsdf.py

Responsible for the TSDF implementation and integration

icp.py

responsible for the pose estimate using ICP whether frame to frame or frame to volume.