Sekundarni povzetek: |
New models and methods have been designed to estimate the influence of the context, the
drivers' activity and behavioural information to the driving style in usual automotive
environment in natural driving and to investigate stress based on smartphone sensors data
considering the current activity. For these purposes, an experiment was conducted using
three types of validation metrics: (i) the stress recognition metric, considering the current
activity based on data collected before driving; (ii) the metric based on a self-assessment of
driving style; (iii) and the metric based on an objective driving data. 67 hours of driving were
collected for further analysis in pilot study. Ten drivers were involved in the experiment.
Algorithms for detecting stress right before driving (based on analysis contextual and
behavioural data) achieved 71.4% of accuracy for the true positive rate using questionnaire
analysis for validation, established by psychologists. The possibility of applying driving style
self-assessments as second validation metric was evaluated as not precise enough. In the
last third metric a new approach was suggested to estimate the driving style based on data
collected before and during the driving tasks including new parameters for data analysis as a
car door opening and closing manner and application for a type activity recognition based on
Google activity recognition API. Further analysis, in which metric of driving style from
objective driving data was correlated with the data collected before driving and with the
data collected before and during the first 1 min of driving, showed significant correlation
results, from 72.7% to 90.9% of true positive rate.
Results of the pilot study for the driving style estimation system showed a success in
recognizing driving style based on the data collected before and during the driving. In cases
when maximal non-invasiveness should be reached, only smartphone and car door as data
sources can be used to estimate the driving style. Considering these demands we were able
to achieve 72.7% of true positive rate of driving style recognition. It is less compared to the
analysis with using the first 1 min of the driving data, but the results are obtained completely
before the driving, so they could be used in advance as feedbacks to the drivers about the
potentially aggressive driving style. The obtained recognition rates lend support to the hypothesis that contextual, behavioural and activity data could be used for the driving style
estimation. |