People flow estimation with radars

Figure 1: Overview of the project

Problem statement

The CORFOU project aims to design a device for measuring anonymised flows of people to assess soft mobility in urban areas and manage security at events. It is based on radar technology: the radar is placed in a street and sends out a very low-power electromagnetic wave, up to 7 times weaker than a Wi-Fi box (EIRP radar = 14 mW), which bounces off groups of people. The radar measures the reflected wave and processes it to produce a “radar map”, in which each person appears as an amplitude peak, i.e. a red dot, as illustrated below.

Figure 2: Radar principle

Contribution

As the only information measured is the reflected wave, radar is totally anonymous by nature, with each person corresponding to a single point on the radar map. The radar map can then be processed by an algorithm to estimate the flow of people, i.e. their number, speed and direction. The research project aims to design the sensor, the appropriate radar signal processing and the flow estimation algorithm.

Figure 3: Measurement prototype

Measurement prototype

The measuring device, in the image above, is a box incorporating the radar and a miniature computer. For the prototyping and development phase, a camera is added above the sensor. It is used as a reference point for the design of the flow estimation algorithm: for each radar measurement, a video measurement is taken by the camera, for which algorithms for estimating the number of people, speed and direction already exist. In this way, the radar's estimation performance can be compared with the camera's. Each video is sent to an ISO27001-certified (information security management standard) processing and storage server. On arrival at the server, the video is anonymized by permanently and irreversibly blurring all characters, and the original video is destroyed on both the camera and the server. The maximum retention time of the original video on the camera is 5 minutes. The processing of the radar signal and camera image is detailed below.

Figure 4: Processing summary

CORFOU_FR