Abstract Summary
Efficient urban governance requires estimating the count of people on the move. This helps maintain real-time safety and deploy any interventions if necessary. For example, depending upon the information regarding the number of pedestrians in a busy market street, the authorities may prohibit vehicular traffic. Furthermore, people count also helps plan future resources towards civic amenities. Presently camera is the widely accepted and deployed solution to perform people counting. However, the wide deployment of cameras in public places is disapproved by people and discouraged by law. Cameras are privacy-invasive and considered susceptible to security incidents. We explore an alternative privacy-aware people counting mechanism through the use of millimeter wave radar as our sensor. Millimeter wave radar senses people as a cluster of points known as point clouds, thus minimizing the data collected by it. This reduces the risk of leakage of personal data during a security breach. Most of the current literature related to mmWave focuses on indoor applications such as gesture recognition and gait detection. These studies have typically been done in a constrained space and under controlled conditions. The indoor studies provide high accuracy but the approaches can not be directly mapped for counting people in an outdoor scenario. Two key challenges that impact the performance of radar-based counting when deployed in an outdoor environment are noise rejection and counting the number of people within a single cluster. Through this work, we attempt to address the aforementioned key challenges by implementing a neural network model to estimate the people count in a point cloud. We deploy a mmWave radar across a street together with a camera to evaluate the efficacy of our proposed solution. The camera images are used as ground truth to train and test the network model. Our initial results indicate that the radar can detect presence with an accuracy greater than 90% and can provide the exact count with an accuracy greater than 80%. Our evaluation shows that mmWave is a promising technology for counting people with inherent privacy awareness.
