Anticipating driver intention is an important task when vehicles of mixed and varying levels of human/machine autonomy share roadways. Driver intention can be leveraged to improve road safety, such as warning surrounding vehicles in the event the driver is attempting a dangerous maneuver. In this work, we propose a novel method of utilizing in-cabin and external camera data to improve state-of-the-art (SOTA) performance in predicting future driver actions. Compared to existing methods, our approach explicitly extracts object and road-level features from external camera data, which we demonstrate are important features for predicting driver intention. Using our handcrafted features as inputs for both a transformer and an LSTM-based architecture, we empirically show that jointly utilizing in-cabin and external features improves performance compared to using in-cabin features alone. Furthermore, our models predict driver maneuvers more accurately and earlier than existing approaches, with an accuracy of 87.5% and an average prediction time of 4.35 seconds before the maneuver takes place. We release our model configurations and training scripts on https://github.com/ykung83/Driver-Intent-Prediction


    Access

    Download


    Export, share and cite



    Title :

    Looking Inside Out: Anticipating Driver Intent From Videos


    Contributors:

    Publication date :

    2023


    Remarks:

    8 pages, 7 figures


    Type of media :

    Preprint


    Type of material :

    Electronic Resource


    Language :

    English




    WARNING DRIVER OF INTENT OF OTHERS

    BENDER MICHAEL / CHILDRESS EDWARD T / CHILDRESS RHONDA L et al. | European Patent Office | 2018

    Free access

    Prediction of driver intent at intersection

    ZELMAN IDO / MUDALIGE UPALI P / ELVITIGALA THANURA RANMAL | European Patent Office | 2019

    Free access

    PREDICTION OF DRIVER INTENT AT INTERSECTION

    ZELMAN IDO / MUDALIGE UPALI P / ELVITIGALA THANURA RANMAL | European Patent Office | 2017

    Free access

    Warning driver of intent of others

    BENDER MICHAEL / CHILDRESS EDWARD T / CHILDRESS RHONDA L et al. | European Patent Office | 2019

    Free access

    Predicting Driver Intent from Models of Naturalistic Driving

    Bender, Asher / Ward, James R. / Worrall, Stewart et al. | IEEE | 2015