Real-Time People Counting System for Customer Movement Analysis

We propose a real-time people-counting system that can be applied in a retail store to estimate the number of people entering and exiting. The proposed method consists of three main procedures: foreground extraction based on the average picture level (APL), a dilated motion search based on the maximum a posteriori probability (MAP), and flow analysis based on multiple touching sections (MTS). We first produce a background model to extract the foreground by using line-wise APL. A dilated motion search with the MAP-based approach is then used to estimate the motions on the line of interest. Next, the flow generated by the foreground on the MTS is analyzed. Finally, the results of the motion estimation and flow analysis are incorporated to produce the number of people entering and exiting the store. We used a low-cost microcomputer to implement the system, which is capable of wireless transmission and is easy to install in a retail environment. Experimental results show that the proposed method provided the best F1 score and accuracy values for the people count results with much lower computational complexity than benchmark methods. In addition, it successfully estimated the number of people entering and exiting the store in real time.