Estimating Crowd Distribution Using Smart Bulbs

TLDR: This work approaches the problem by analyzing the Wi-Fi packets for counting people and estimating their position within a pre-defined accuracy, and proposes improvised counting techniques that results in people counts close to 75% of the ground truth.

Abstract: Many IoT applications require the knowledge of crowd distribution, particularly in indoor scenarios. To this end, we leverage the lighting grid infrastructure in buildings by using smart light bulbs, which can include variety of sensors, in these grids. The exponentially increasing adoption of smartphones and the Wi-Fi infrastructure has motivated us to tackle this problem using Wi-Fi. As we seek a solution that works in many buildings, relying on active user participation or installing apps is not an option. Therefore, we need a Wi-Fi based crowd distribution estimation technique that is non-participatory and non-intrusive, and works with very few Wi-Fi packets generated by users' smartphones sporadically. We approach the problem by analyzing the Wi-Fi packets for counting people (smartphones) and estimating their position within a pre-defined accuracy. To this end, extensive experiments are conducted in a real-world testbed with controlled settings as well as in test setups in office spaces with no control. We propose improvised counting techniques that results in people counts close to 75% of the ground truth. We further propose improvements to range-free localization techniques to refine the position estimation accuracy and reduce the execution time. Our algorithm estimates the location with an accuracy of 2m 74% of the time, when Wi-Fi sniffers are placed in bulbs every 4m in the grid.