Objective To construct a risk assessment model based on healthcare workers’ (HCWs) real behavioral data, and precisely quantify virus exposure dynamics and infection risks of HCWs during routine clinical practice. Methods HCWs and patients from otolaryngology outpatient clinics at a general tertiary first-class hospital in Beijing were selected as the research subjects. Behavior information of relevant individuals including relative position, interpersonal distance, facial orientation, and contact frequency, were recorded and stored using depth cameras, laptops, portable power supplies, tripod stands, and removable hard drives. Behavior data were processed by semi-supervised machine learning algorithms. A mathematical model was finally constructed to assess real-time viral exposure and infection risks during HCWs’ procedure. Results A total of 161 917 behavioral datasets were captured in this study, including external auditory canal irrigation, nasal/laryngeal/otoscopic endoscopies. After inspection and exclusion, 103 788 behavioral datasets remained. HCWs in otolaryngology outpatient spent 90.61% of clinical time in close contact on average, predominantly in face-to-side (F-S) orientation. Nasal endoscopy generated peak viral RNA loads, ranging from 0.543 to 0.745 viral RNA loads/s, with an overall infection risk of 42.74%. Infection risks could be reduced to 2.44%, 14.20%, and 36.05% when wearing N95, medical surgical, and cotton masks, respectively. Conclusion HCWs in otolaryngology outpatient clinics face high risks of occupational exposure in their daily work. Personal protective equipment should be used strictly, and appropriate interpersonal distance and optimal face orientation should be maintained.