Human Flow Measurement System Using Floor Estimation of Depth Images for Low-End IoT Devices
Takuya Nagatoshi, Michiharu Niimi · 2023
We propose a system that uses depth information, which represents the distance from the sensor, instead of color information to do both measure human flow and protect privacy for low-end IoT devices. The system is designed to detect the position and number of persons from depth information. Since administrative organizations or educational institutions have been reducing their budgets, this system should be implemented at as low a cost as possible. In order to realize human flow detection system on low-end Iot devices, we use low performance depth cameras as data acquisition device controlled by single board computers, such as Raspberry Pi. As one of our goals is all data processing are performed on single board computers, we adopt computational methods for the detection as possible as simple. The proposed method is based on the background subtraction method, which prepares a reference depth image and extracts moving regions from one frame extracted from the video depth image. Furthermore, we aim to achieve high-precision people flow measurement by combining the following three elements: segmentation of moving objects using edge information, identification of human areas using floor information, and human tracking using areas where people overlap in the direction of the time axis. Experiments were also conducted and evaluated in a real space using a program that implements the presented method.