Real-Time Privacy Preserving Crowd Estimation Based on Sensor Data

As one of the popular topics to ensure public safety, crowd estimation has attracted lots of attentions from both industry and academia. Most of traditional crowd estimation approaches rely on sophisticated computer vision algorithms to estimate crowd based on camera data, therefore suffering from privacy issues and high deployment and data processing cost. In this paper we present a sensor fusion based approach to real-time crowd estimation based on privacy-conscious and inexpensive sensors. The approach has been implemented and verified first by a small scale deployment at our lab, and then tested based on a 3-month trial at a shopping mall in Singapore. A deep analysis has been carried out based on the data sets collected from the trial, showing promising results: 1) the data from CO2, sound pressure and infrared sensors are influential in estimating crowd levels for indoor environments, 2) Random Forest and C4.5 are identified as the more suitable supervised learning models, 3) an accuracy of 95% can be achieved by our crowd estimation system in a real scenario. In contrast to the state of the art, our approach is privacy preserving and can provide comparable estimation accuracy with lower deployment and processing cost and better applicability for large scale setups. It can be used either as an alternative solution when user privacy must be enforced or as a complementary solution to camera-based crowd estimation when privacy is less concerned because of pubic safety.