gleisurfer

What is it ?

Gleisurfer is an iOS app that helps the railway workers get a no-fuss clearance outline and anomalous object detection using augmented reality and machine learning.

Why?

The clearance outlining on the railroad tracks are hard to do manually with bulky spatial calliper. Today smartphones are collecting way more data than we need! We can use state-of-the-art point cloud generation, 3D reconstruction and segmentation methods to safely create the outline of the train clearance and detect if an object is not supposed to be there!

Rail marking

Sample input	Sample output

References

This code use the rail_marking repo with a few changes in test_video.py and test_one_image.py under scripts/segmentation. It also uses yolov5 for object detection.

Environment

conda create --name gleisurfer python=3.8
conda activate gleisurfer
conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=10.2 -c pytorch
conda install -c conda-forge brotlipy
pip install matplotlib requests albumentations==0.4.3 nudged tqdm pandas seaborn psutil

Usage

Pretrained segmentation model needs to be downloaded in the folder.

Here is an example usage of the script with one of the sample videos

python scripts/segmentation/test_video.py -snapshot bisenetv2_checkpoint_BiSeNetV2_epoch_300.pth -video_path /path_to_videos/Videos_with_sensordata/2022_08_17_14_00_27/movie.mp4