its_research

We are developing possible solutions to solve problems in the context of Intelligent Transportation System. More precisely, labeling problems and anomaly detection on bus trajectory (GPS).

Disclaimer

The preprocessed data from Dublin is available at: https://drive.google.com/file/d/1tkpxtFulyWQhcaRuCqsDVMBLWW8UN3LL/view?usp=sharing https://drive.google.com/file/d/1FmjM2Xi-mbwZALOTQcHBEgmg7uc71zwq/view?usp=sharing

All Recife data is private.

The dataset can be used to reproduce the experiments.

Point Activity Classification

It is a stacked deep-learning model composed of recurrent and attention layers that learn a vector representation for each trajectory point by classifying it into a set of activity points (in route, bus stop, traffic signal, and other stops) based on temporal and spatial features.

A Spatial-Temporal Outlier Detection

We aim to detect anomalous bus trajectories using supervised learning. Thus, we propose a multi-class classifier that learns the typical behavior of buses but, instead of performing a hard detection decision as literature approaches, ou solution calculates an anomaly score based on the uncertainty of the classifier.

Applying A Transformer Language Model for Anomaly Detection in Bus Trajectories

• Contents:
  • process_dublin.py
  • pipeline_to_generate_data.py
  • preprocess_recife.py
  • run_loop_transformer.py
  • transformer_model.py

The preprocessed data to test our solution is available at: https://drive.google.com/file/d/12cDmUdY5lDEcLzBDaPtAGzs66MokqfuY/view?usp=sharing

Contents

• Requirements:
• library versions:
  • numpy = 1.18.1
  • h5py = 1.10.4
  • tensorflow = 2.1.0
  • h3 = 3.4.3
  • gmplot = 1.2.0
  • scikit-learn = 0.22.2
  • scipy = 1.4.1
  • gensim = 3.8.0
  • keras = 2.3.1
  • plotly = 4.5.4
• Models