Challenges for Driver Action Recognition with Face Masks

Advanced Driver Assistance Systems (ADAS) are enabling technologies in Intelligent Transportation Systems. Modern ADAS include algorithms to classify drivers' actions and distractions, aiming at identifying situations in which the driver is inattentive. Such systems typically include components for Driver Action Recognition (DAR) and Visual Distraction Classification (VDC), which prevent risky situations during semi-autonomous driving. DAR and VDC often rely on cameras that track the driver and classify actions based on image recognition algorithms. The COVID-19 pandemic has changed several common social behaviours, including the widespread use of face mask even during driving. In some cases (taxi, bus) face covering policies are compulsory in many legislations. We here show that these behavioural changes challenge state-of-theart DAR and VDC systems, with the average F1-score in some scenarios dropping by around 30% when exposed to images of drivers wearing masks. Noting a lack of public datasets to update the ML classifiers performing such tasks, we contribute MASKDAR, a dataset for Action Recognition of Drivers wearing face Masks. Finally, using MASKDAR we show the importance of including subjects with face masks in datasets for DAR.

Challenges for Driver Action Recognition with Face Masks

Advanced Driver Assistance Systems (ADAS) are enabling technologies in Intelligent Transportation Systems. Modern ADAS include algorithms to classify drivers' actions and distractions, aiming at identifying situations in which the driver is inattentive. Such systems typically include components for Driver Action Recognition (DAR) and Visual Distraction Classification (VDC), which prevent risky situations during semi-autonomous driving. DAR and VDC often rely on cameras that track the driver and classify actions based on image recognition algorithms. The COVID-19 pandemic has changed several common social behaviours, including the widespread use of face mask even during driving. In some cases (taxi, bus) face covering policies are compulsory in many legislations. We here show that these behavioural changes challenge state-of-the-art DAR and VDC systems, with the average F1-score in some scenarios dropping by around 30% when exposed to images of drivers wearing masks. Noting a lack of public datasets to update the ML classifiers performing such tasks, we contribute Maskdar, a dataset for Action Recognition of Drivers wearing face Masks. Finally, using Maskdarwe show the importance of including subjects with face masks in datasets for DAR. © 2022 IEEE.

Continuity of care in a psychiatric Day Hospital during the COVID-19 emergency. Transformations of therapeutic models between crisis and change

Introduction: The COVID-19 pandemic has challenged our treatment models, leading our team to rethink and share strategies of care, in order to ensure the continuity of therapeutic relationships during the precarious context of the pandemic. Objective: To evaluate the feasibility and perceived efficacy of a remote therapeutic intervention in the treatment of patients in a psychiatric Day Hospital (DH). Methods: A group of 48 children and adolescents (64.6% female, age from 9 to 19 years, mean age 15.2 years and SD 2.50), already in treatment for severe and complex psychopathology at the psychiatric-therapeutic DH of the Child Neuropsychiatry Service of the Regina Margherita Hospital of Turin prior to the COVID-19 emergency, received remotely a multidisciplinary intervention for 90 days, with maintenance of the weekly / multi-weekly therapeutic session and additional contacts based on individual needs. Patients were prospectively assessed, at the lockdown onset (T0) and after 30 (T1), 60 (T2) and 90 days (T3), with clinical evaluation of the overall severity of psychopathology using the Clinical Global Impressions-Severity scales (CGI-S) and -Improvement (CGI-I) and monitoring suicidal ideation or behavior, non-suicidal self-harm (NSSI) and social withdrawal. Patients and parents were asked to indicate on a Likert scale the perceived usefulness of the telemedicine interventions. Data were analyzed with non-parametric tests and an ordinal logistic regression analysis was performed. Results: An average of 5.4 remote interventions per patient per week (min: 2, max: 9) were provided, without any complete drop-out. During the 90 days of the intervention, no suicide attempt was recorded. Three patients (6.2%), were hospitalized. CGI-S scores indicated a significant reduction in the severity of illness between T0 and T3 (p = 0.03), with improvement in 45.8% of patients, worsening in 25%, and no change in the remaining 29.2%. Clinical improvement was more evident among patients without social withdrawal than in those with social withdrawal (p 0.04). Suicidal thoughts were reported by 35.4% of the participants at T0 and 20.8% at T3. Based on information from patients, parents and educators, NSSI was present in 22.9% at T0 and in 12.5% at T3 (p = 0.063). Of the patients with NSSI at T0, 45.4% showed a reduction of self-injury. There were no newly reported cases of NSSI during the 90 days of intervention. Individual NPI visits were rated from moderately to extremely useful by 92.3% of the patients, and 82% of the parents rated from moderately to extremely useful NPI visits, psychotherapy sessions and teaching. Conclusions: The remote intervention during the lockdown period COVID-19 was feasible, well accepted by patients and parents and accompanied by indicators of improvement, despite the necessary suspension of in person DH treatments. It is conceivable to hypothesize that the continuity of care through telemedicine and the flexibility of the multidisciplinary model helped patients maintain the therapeutic alliance. (PsycInfo Database Record (c) 2021 APA, all rights reserved) Abstract (Italian) Introduzione: La pandemia da COVID-19 ha messo alla prova i nostri modelli di cura lanciando nello stesso tempo una sfida: ripensarli e condividerli, per garantire continuita di relazione terapeutica pur in un contesto di emergenza e precarieta. Obiettivo: Valutare la fattibilita e efficacia percepita di un intervento terapeutico da remoto nel trattamento di pazienti in un Day Hospital psichiatrico (DH). Metodi: Un gruppo di 48 bambini e adolescenti (64,6% di sesso femminile, eta da 9 a 19 anni, eta media 15,2 e DS 2,50), gia in trattamento per psicopatologia grave e complessa presso il DH psichiatrico-terapeutico del Servizio di Neuropsichiatria Infantile dell'Ospedale Regina Margherita di Torino nel periodo precedente all'emergenza sanitaria, hanno ricevuto un intervento multidisciplinare da remoto per 90 giorni: mantenimento dei setting terapeutici settimanali/plurisettimanali e contatti in urgenza in base ai bisogni individuali dei pa ienti. I pazienti sono stati seguiti nel tempo, prima dell'intervento (T0) e dopo 30 (T1), 60 (T2), e 90 giorni (T3), con valutazione clinica della gravita complessiva della psicopatologia mediante le scale Clinical Global Impressions-Severity (CGI-S) e -Improvement (CGI-I) e monitoraggio di possibile ideazione o comportamento suicidari, autolesionismo non suicidario (NSSI), e ritiro sociale. A pazienti e genitori e stato chiesto di indicare su scala Likert l'utilita percepita degli interventi erogati in modalita telematica. I dati sono stati analizzati con test non parametrici ed e stata effettuata un'analisi di regressione logistica ordinale. Risultati: Gli interventi erogati a distanza sono stati in media 5,4 per paziente per settimana (min: 2;max: 9). Nessun paziente ha abbandonato il trattamento. Non si sono verificati drop out completi rispetto alla presa in carico. Non si sono verificati tentativi di suicidio. Per 3 pazienti (6.2%) si e reso necessario programmare un ricovero. La CGI-S ha indicato una significativa riduzione della severita del quadro clinico (p = 0,03), con un miglioramento nel 45,8% dei pazienti, una stazionarieta nel 29,2% e peggioramento nel 25%. Il miglioramento clinico si e reso piu evidente nel gruppo di pazienti senza ritiro sociale rispetto al gruppo di pazienti con ritiro sociale (p = 0,04). Ideazione suicidaria era presente nel 35,4% dei pazienti a T0 e nel 20,8% a T3. Sulla base delle dichiarazioni dei pazienti, dei genitori e degli educatori, i pazienti con NSSI a T0 erano il 22,9% e a T3 il 12,5% (p = 0,063). Tra i pazienti con NSSI a T0 si e riscontrata una riduzione del NSSI a T3. Non si sono registrati nuovi casi di NSSI. Le visite NPI sono state valutate da moderatamente a estremamente utili dal 92,3% dei pazienti e l'82% dei genitori ha valutato da moderatamente a estremamente utili le visite NPI, le sedute di psicoterapia e la didattica a distanza. Conclusioni: L'intervento a distanza durante il periodo di lockdown COVID-19 si e dimostrato fattibile, ben accettato da pazienti e genitori, e accompagnato da indici di miglioramento, nonostante la necessaria sospensione dei trattamenti in presenza. E ipotizzabile che la continuita del trattamento e la flessibilita del modello multidisciplinare, anche in modalita telematica, abbiano consentito ai pazienti di mantenere l'alleanza terapeutica e di conservare la fiducia nella relazione di cura. (PsycInfo Database Record (c) 2021 APA, all rights reserved)