maneuverRecognition -- A Python package for Timeseries Classification in the domain of Vehicle Telematics

TL;DR

The paper introduces maneuverRecognition, a Python package designed for time series classification in vehicle telematics, facilitating data preprocessing, model building, and evaluation for driving maneuver recognition using real smartphone sensor data.

Contribution

It provides a practical Python toolkit with ready-to-use models and functions tailored for maneuver recognition in vehicle telematics, addressing a gap in accessible tools for researchers and practitioners.

Findings

Successfully applied to real driving data from three individuals.

Includes a modifiable LSTM-based network structure.

Enhances data processing and model evaluation workflows.

Abstract

In the domain of vehicle telematics the automated recognition of driving maneuvers is used to classify and evaluate driving behaviour. This not only serves as a component to enhance the personalization of insurance policies, but also to increase road safety, reduce accidents and the associated costs as well as to reduce fuel consumption and support environmentally friendly driving. In this context maneuver recognition technically requires a continuous application of time series classification which poses special challenges to the transfer, preprocessing and storage of telematic sensor data, the training of predictive models, and the prediction itself. Although much research has been done in the field of gathering relevant data or regarding the methods to build predictive models for the task of maneuver recognition, there is a practical need for python packages and functions that allow to quickly transform data into the required structure as well as to build and evaluate such models. The maneuverRecognition package was therefore developed to provide the necessary functions for preprocessing, modelling and evaluation and also includes a ready to use LSTM based network structure that can be modified. The implementation of the package is demonstrated using real driving data of three different persons recorded via smartphone sensors.