Crowd control and public safety are far more seriously threatened by large crowds at events like political rallies, concerts, and religious festivals. This paper proposes an Al-driven real-time crowd monitoring system that integrates multiple advanced technologies to enhance situational awareness and security enforcement. The proposed system employs Social Long Short-Term Memory (LSTM) networks for precise trajectory prediction, deep learning-based facial recognition utilizing Facial Expression Recognition (FER) to facilitate the rapid identification of individuals, and ResNet-enhanced group emotion analysis to assess crowd sentiment and identify distress signals through gesture recognition. In addition, the system is backed by a permissioned blockchain network, which ensures secure and tamper-proof data recording. A case study in the city of Varanasi, India, demonstrates the practical applicability of the system, particularly in the management of large crowds during major festivals like Dev Diwali and Kartik Purnima. The framework enhances the capacity of law enforcement agencies and event planners to respond rapidly to emergencies, more effectively manage crowd dynamics, and reduce criminal activities such as harassment and theft, while capturing real-time audience sentiment. Experimental validation demonstrates that the system achieves 92.5% accuracy in emotion detection and 90% accuracy in facial recognition, underscoring its viability for large-scale deployment. A more nuanced knowledge of crowd sentiment is made possible by emotion and gesture recognition, which encourages taking preventative actions before things get worse. The fusion of Al analytics with blockchain transparency fosters proactive threat detection and refined crowd sentiment assessment, which significantly advances intelligent surveillance, security automation, and public safety technologies.