Recognizing Flu-like Symptoms from Videos




Brief Description

While the recent swine flu pandemic was luckily less severe than initially thought, there remains a constant threat of mutated or reassorted influenza strains that give rise to new outbreaks that could range from small local clusters to seasonal epidemics or even global pandemics. Similarly, history has also shown us that previously unknown pathogens such as the SARS coronavirus could emerge and cause serious outbreaks. Respiratory diseases often manifest themselves through similar flu-like symptoms and early detection of new outbreaks is of central importance in order to delay or prevent their escalation and wider spread. However, classical surveillance systems are mostly relying on time-delayed and costly virological tests requiring hospital or physician visits.


One potential alternative is to detect typical flu-like symptom in human behaviors, by automatically analyzing video footage from public areas such as airports, bus stations, which exploits the existing vision-based surveillance infrastructure in public venues. This will provide a unique valuable source of information that is complementary to the existing public health monitoring network. Under this context, we make a first attempt on the recognition of typical flu-like symptoms: sneeze and cough actions. This gives rise to a human action video dataset dedicated towards the problem of flu-like symptoms recognition and detection, which is of central importance in early surveillance of respiratory disease outbreaks.

For this purpose, a dedicated video dataset has been created, referred to as the BII Sneeze-Cough Human Action Video Dataset, or BIISC. Cropped snapshots of this dataset are presented below:


From top to down shows eight actions: answer phone call, cough, drink, scratch head, sneeze, stretch arm, wave hand and wipe glasses. From left to right shows six pose-and-view variations: stand-front, stand-left, stand-right, walk-front, walk-left, and walk-right.



BII Sneeze-Cough Human Action Video Dataset: (readme)





T. Thi, L. Cheng, L. Wang, N. Ye, J. Zhang, and S. Maurer-Stroh. Recognizing Flu-like Symptoms from VideosBMC Bioinformatics, 2014