Towards Efficient Detection and Crowd Management for Law Enforcing Agencies

Abstract

each year more than two million Muslims from around the world come to perform Hajj in Makah. It is considered the world's largest recorded human gathering during any worshiping event. Safety makes one of the main concerns with regards to managing such large crowds for ensuring that stampedes and other similar overcrowding accidents are avoided. For this purpose, 5000 cameras are installed around the holy sites for monitoring purposes. Due to the continuous nature of surveillance systems in generating video data, it is almost impossible to efficiently and accurately monitor an event of this size in real-time. Analyzing such huge data has required a lot of human resources. Therefore, there is a great need for advanced intelligent techniques to automatically count and manage such large crowds. In order to create an advanced intelligent system that contributes to crowds counting and managing through the surveillance system. In this paper, we propose an accurate computer vision-based approach to crowd management using Convolutional Neural Network (CNN). Our proposed framework is three folds. In the first fold, our own dataset for pilgrim detection is created, covering both sparse and dense crowds. In the second fold, a Faster-RCNN object detection model is trained to detect and count the number of pilgrims. In the third fold, utilizing the resources efficiently the surveillance system has used frame differencing technique to differentiate between motion and static video frames. Only in the case of some sort of motion, we will pass these frames to the pilgrims counting model to tell us about the number of pilgrims in the video. When the number of pilgrims counting is exceeded from the pre-defined threshold the system will automatically trigger the alarm pointing the camera to the location to inform the concerned authorities to take action appropriate measures. Along with that, only the dense crowd will be monitored by law enforcement and for better management. Our experiments show that Faster Region CNN (Faster RCNN) is suitable for accurate detection when compared with other state-of-art crowd management techniques so far reported.