A real-time intelligent monitoring method for indoor evacuee distribution based on deep learning and spatial division

Abstract

Providing real-time and detailed indoor evacuee distribution information is of great significance for the on-site formulating and dynamic adjustment of indoor emergency evacuation strategies. However, due to limitations in hardware costs and privacy protection, existing video surveillance and other evacuee counting methods cannot achieve real-time and accurate monitoring of the number and distribution of evacuees in all indoor areas. To solve the problem, using video data from an indoor surveillance system as a data source, an intelligent monitoring method for indoor pedestrian real-time distribution based on deep learning and spatial division is proposed. Firstly, the whole indoor area is divided many subareas according to the layout of indoor space and the size of spatial units. Then, deep learning detection and tracking algorithms are used to detect and track evacuees to achieve the number of evacuees at the boundary of each subarea and their movement directions; Finally, the numbers of evacuees entering and leaving each subarea are counted to obtain the spatial distribution information of evacuees. The experimental results show that within an appropriate monitoring distance, the proposed method achieves an average F1 score of 91.85% for evacuee counting at the boundaries of each subarea, with a processing speed of 22 FPS. More comprehensive and accurate real-time monitoring of evacuee distribution in the entire indoor space can be achieved, including no covered areas of cameras. It not only helps to formulate on-site emergency evacuation in case of a fire, but also enhances the daily operation and management capabilities of buildings.