A Human Face Detector for Big Data Analysis of Pilgrim Flow Rates in Hajj and Umrah

Abstract

In today's digital world, some crowded venues still rely on outdated methods, such as counting people using counters or sensors at the entrance. These techniques generally fail in areas where people move randomly. Crowd management is an important challenge for ensuring human safety. This paper focuses on developing a crowd management system for Hajj and Umrah duty. Motivated by the recent artificial intelligence techniques and the availability of large-scale data, a crowd management system was established and is presented in this paper. Utilizing the most recent Deep Learning techniques, the proposed crowd management system will be charged with detecting human faces, face identification, tracking, and human face counting tasks. Face counting and detection will be achieved by computing the number of people in a given area. Face detection and tracking will be carried out for person identification, flow rate estimation, and security. The suggested crowd management system is composed of three key components: (1) face detection, (2) assignment of a specific identifier (ID) to each detected face, (3) each detected face will be compared to the stored faces in the dataset. If the detected face is identified, it will be assigned to its ID, or a new ID will be assigned. The crowd management system has been developed to improve the Cross-Stage Partial Network (CSPNet) with attention module integration. An attention module was employed to address object location challenges and a channel-wise attention module for determining the objects of focus. Extensive experiments on the WIDER FACE dataset proved the robustness of the proposed face detection module, which allows for building reliable crowd management and flow rate estimation systems through detecting, tracking, and counting human faces. The reported results demonstrated the power of the proposed method while achieving high detection performance in terms of processing speed and detection accuracy.