Multi-Sensor-Fusion System for People Counting Applications

Abstract

For economical as well as environmental reasons, logistical planning and efficient assignment of transport vehicles in public transportation need a precise knowledge of passenger utilization. To provide reliable figures sophisticated counting methods are demanded. Previously developed systems, mostly using a single 2D image sensor or a 3D depth sensor, can not fully achieve the required accuracy. In this paper, we present a robust people counting algorithm, based on multi sensor data fusion. Our solution runs on embedded systems with reasonable requirements with respect to computational power. 3D distance information, obtained from the ToF system, is used to perform the basic detection of objects. Once detected, the objects get additional classification features exploiting the data provided by a RGB camera and an IR thermal sensor. We describe the current state of fusing and processing of the collected data including the detection, classification (vital, non-vital), as well as sequential tracking and counting. We provide current counting results along with insights to future development concepts to improve the stated algorithms especially in terms of vital, non-vital classification and object recognition.