Real-time personnel counting of indoor area division based on improved YOLOV4-Tiny

Abstract

With the growth of building energy consumption, determining how to effectively reduce energy waste has become a matter of concern. We believe that by acquiring the basic data for intelligent regulation such as the location and number of personnel, we can achieve the intelligent adjustment of indoor energy-using equipment and ultimately reduce energy consumption. First, this study describes the division of an indoor scene by space coordinate transformation. Second, considering the complex indoor scenes, the main module of the backbone feature network is improved based on YOLOV4-Tiny, making the algorithm more suitable for people detection in complex indoor environments. Finally, the improved algorithm adopted in this study is verified in the indoor real-time monitoring scene. The experimental results show that the improved network's ability to detect people in complex indoor scenes is improved. The overall average accuracy is increased by 18.15%, and the detection speed reaches about 73 Frames Per Second (FPS).