Counting and detecting people with radars. : The comparison of three radars in relation to their people counting ability.

Abstract: People counting and human detection systems have during the last decade been developed using many different sensors; from video cameras to radars. While video cameras function much like the human eyes and yield data relatable to humans, it might be prone to privacy regulations. Therefore, sensors not affected by these privacy regulations have been used to develop people counting systems. In this thesis, radar sensors are examined in their people counting ability. The radars in question are: InnoSent iSYS5005, RFbeam K-LD7, and Acconeer XM112. An investigation of the three different radars as well as different geometries is carried out in order to find the most suitable radar and the optimal positioning for people counting. This thesis follows the principle of Design Science Research, where a controlled experiment was conducted in order to answer the research questions. We look at accuracy and precision together with other variables to determine which radar and what geometry is the most promising. In the end, we find that the iSYS5005 has the most promising result. However, the results were not as promising as one could have hoped, showing an overall poor performance. We argue that this is because the radar regularly loses track of objects, lowering the overall accuracy and precision. Which in turn lowers the overall performance of the sensors. Therefore, we mean that this shows the importance of post processing algorithms to interpret the data. We propose a tardiness probability algorithm which we believe could fix many of the issues the unprocessed data has.