StereoDemo

Demo accompanying the article "RADIUS: Robust Anomaly Detection in Urban Drainage with Stereovision"

Requires python3.6 or newer.

Install necessary packages:

python3 -m pip install -r requirements.txt --user

Run the demo:

python3 demo.py

After choosing a data set to run on, the script will run the first three steps of the framework, "Data Acquisition", "Semi-Global Stereo Matching", and "Three-Dimensional Geometry Reconstruction".

At this point the user will need to select a valid Z-range in the pptk pointcloud viewer, using instructions as specified in the terminal. It's important to select a wide enough range to capture the overal shape of the pipe but not include too much points outside the pipe. Values around -1.5 and -2.0 are suggested for most image sets.

A RANSAC procedure will then commence to fit the data within the selected Z-range to the model. We perform 10 iterations of RANSAC, repeated a maximum of three times if no suitable models are found the first 2 times.

Then the anomaly scores will be mapped onto the point cloud, the visualisation of the model, and the image. Output files will be written to the "output/" folder for closer inspection.

To navigate the pptk point cloud viewer, use:

• Click and drag to rotate
• Shift-click and drag to pan
• Scroll to zoom
• 1, 3, 7 keys to align the view with the X, Y, Z axes respectively
• 5 key to switch between perspective and orthographic view
• [ and ] keys to change the color attribute between anomaly score and rgb pixel color (only after model fit)