English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Driver observation models are rarely deployed under perfect conditions. Inpractice, illumination, camera placement and type differ from the ones presentduring training and unforeseen behaviours may occur at any time. Whileobserving the human behind the steering wheel leads to more intuitivehuman-vehicle-interaction and safer driving, it requires recognition algorithmswhich do not only predict the correct driver state, but also determine theirprediction quality through realistic and interpretable confidence measures.Reliable uncertainty estimates are crucial for building trust and are a seriousobstacle for deploying activity recognition networks in real driving systems.In this work, we for the first time examine how well the confidence values ofmodern driver observation models indeed match the probability of the correctoutcome and show that raw neural network-based approaches tend to significantlyoverestimate their prediction quality. To correct this misalignment between theconfidence values and the actual uncertainty, we consider two strategies.First, we enhance two activity recognition models often used for driverobservation with temperature scaling-an off-the-shelf method for confidencecalibration in image classification. Then, we introduce Calibrated ActionRecognition with Input Guidance (CARING)-a novel approach leveraging anadditional neural network to learn scaling the confidences depending on thevideo representation. Extensive experiments on the Drive&Act datasetdemonstrate that both strategies drastically improve the quality of modelconfidences, while our CARING model out-performs both, the originalarchitectures and their temperature scaling enhancement, leading to bestuncertainty estimates.

Is my Driver Observation Model Overconfident? Input-guided Calibration Networks for Reliable and Interpretable Confidence Estimates