| Summary: | With more countries entering the endemic phase of COVID-19 amid increasingly contagious COVID-19 mutations, the risk of exposure to the virus is greater than ever. Safe distancing and limiting crowds have been the cornerstone to contain the community spread of the virus. Hence, there is a need for a crowd management system to provide crowd information of places so the public can make informed decisions on places to visit.
A crowdsourced image-based system was developed recently to overcome the limitations of other approaches such as cost. However, the system relies heavily on user inputs to generate crowd count and historical data to perform forecasts for locations, which often results in scarce information as the platform suffers from low take-up and lack of upload.
A key challenge for crowdsourcing is a lack of incentive for users to contribute. Therefore, in this project, a reinforcement learning-based dynamic incentive mechanism was introduced to optimally allocate reward points to maximize user uploads and encourage user contributions. Google Popular Times data was also incorporated to display bar charts and heatmap, allowing users to view hourly and weekly crowd trends of the location to provide better decision support.
Simulation results comparing the proposed dynamic incentive scheme against static schemes concluded that the proposed scheme was better overall as it adapts to changing sensing demand and upload supply, was forward-looking by considering future crowd level and is budget aware.
|
|---|