Driver Behaviour Study/Analysis with DL

This repository is dedicated to data analysis on time-series provided by CAN data from a vehicle on the road.

Data

Data comes from: https://github.com/Eromera/uah_driveset_reader
Some information about the column names is avaialble:

• https://github.com/Eromera/uah_driveset_reader/blob/master/driveset_reader.py
• https://www.researchgate.net/publication/307628008_Need_Data_for_Driver_Behaviour_Analysis_Presenting_the_Public_UAH-DriveSet

Analysis

The repo is organized as follows:

• lib: a library with all useful functions for this study but could be re-used for more
• notebooks: the notebooks used to investigate the data and experiment. They are organized by their prefix "X_". If there is no prefix it is either old or something out of ordinary

Inspiration and help

This is heavily inspired (with copy/paste) from the following article: Data Exploration with Adversarial Autoencoders

To do(s)

I will come back to it to:

• check yaw rate values
• improve GAN clustering
• test with more features such as OpenStreetMap or PROC VEHICLE DETECTION

Conclusion

Notebook:

Notebook link

Notebook Github link

Plots:

• GAN output link
  
• Andrew curves of the encoder features
  
• T-SNE link
  
• Line plots per cluster link
  
• Trajectories per cluster link